US 20050141603 A1 Abstract A method and modem for communicating serial input data over a transmission link. Serial input data is partitioned into parallel data elements prior to rotation by an invertible linear mapping. Resulting frames of parallel signal elements sequentially modulate a carrier, which is then transmitted over the link. After receipt of the modulated carrier from the link, the signal is demodulated and assembled into frames of parallel signal elements which are de-rotated by an inverse linear mapping. Thresholding the result of the inverse mapping recovers the parallel data elements, which are then re-assembled into serial output data. The linear mapping employs: 1) commuting rotation matrices for convolutionally rotating data vectors into signal vectors and vice-versa; 2) filter bank polyphase rotation matrices; or 3) computationally efficient multi-rate wavelet filter banks. Transmitter pre-emphasis places most of the information in lower baseband frequencies; complimentary de-emphasis occurs in the receiver. Logarithmic amplification of the baseband signal prior to carrier modulation improves modulation gain and transmit channel noise attenuation. Coefficients of the rotation matrix of the receiver are adaptively equalized to correct for transmission path distortion. FM double-side band is employed in systems requiring minimized cost and complexity. FM single-side band is employed in systems in which bandwidth reduction is desirable. AM is also employable.
Claims(1) 1. A system for transmitting a first data stream over a transmission medium in conjunction with a second signal, comprising:
a first coder for receiving the first data stream on a plurality of channels thereof, for coding the first data stream, and for providing parallel, coded output data samples; a first synthesizer, in communication with the first coder, for assembling the parallel, coded output data samples into sequences of data frames each comprising a plurality of data coordinates, for converting each frame of data coordinates into orthogonal frames of signal coordinates using an invertible transformation, and for sequentially providing the signal coordinates as a first signal representation of the first data stream; and a summing circuit, in communication with the first synthesizer, for combining the first signal representation of the first data stream with the second signal and for coupling the resulting combination signal to the transmission medium. Description This application is a continuation-in-part of application Ser. No. 08/342,259, filed Nov. 18, 1994 (Attorney Docket MILL-001 AX), which is a continuation of U.S. Pat. No. 5,367,516, filed Mar. 17, 1993, issued Nov. 22, 1994 (Attorney Docket MILL-001XX). This invention relates to signal transmitting and receiving systems and more particularly to modems employing polyphase filtering techniques. As is known in the art, a modulator-demodulator (modem) is an electronic device that modulates transmitted signals and demodulates received signals. The modem generally provides an interface between digital devices and an analog communications system to thus make possible analog transmission of digital information between two terminals or stations. Such transmissions may be over a transmission link such as a telephone line, cellular communication link, satellite link, and cable TV, each of which being generally band-limited. That is, the information may be transmitted across the transmission link only over a predetermined range of frequencies having a maximum bit error rate. As is also known, a modem is used to provide wireless transmission between transmitting and receiving stations. Such wireless communication can be employed in a variety of applications: VHF, IS-54 (cellular), IS-95 (cellular), SPADE (satellite), GSM (cellular), HDTV, and SAT-TV, each using one of the following linear modulation techniques: QAM; QPSK; Pi/4 DPSK; GMSK. As with the transmission line application, the bandwidth of each is limited within an acceptable bit error rate. While most modems are capable of providing compensation for Guassian noise, impulse noise is not well managed. Most modems also require higher powered amplification means since, in all cases, amplitude distortion is unacceptable. In broadcast environments, it is well understood that FM transmission provides superior impulse noise handling. Also in FM transmission schemes, the transmitted signal may be amplified with almost 100% efficiency since the information carrying portion of the signal is identified by the zero-crossings of the signal. Thus, amplitude distortion is ignored. Despite these apparent advantages, only a few isolated examples exist of nonlinear frequency modulation used within modems. Frequency shift keying, providing a 300 bits per second data rate, essentially utilized two frequencies, each for representing a separate data bit. The next major development in modems was the use of phase modulation, beginning with two phase modulation, then four, then eight. A combination of amplitude and phase modulation was later developed, also referred to as quadrature amplitude modulation or QAM. A subsequent development was Guassian minimum shift keying, or GMSK. At first glance, such coding resembles four phase modulation, though in order to avoid amplitude modulation, a special low pass filter referred to as a Guassian filter was applied to the data going into the phase modulator. Since some have regarded the Guassian filter as akin to an integrator, an argument could be made that such a modulation scheme results in frequency modulation, since GMSK can be demodulated with an FM discriminator. Yet, GMSK applies linear functions of the data to in-phase and quadrature carriers to produce a linear modulation; whereas, true FM is mathematically equivalent to the application of trigonometric functions of the filtered signal to an in-phase and quadrature carrier. FM is a non-linear modulation, which was perceived as inefficient for data transmission since the transmitted frequency spectrum is not a simple translation of the baseband spectrum as in AM modems. A further perceived problem with FM for a modem is that it only accepts real signals into its voltage controlled oscillator. That is, in the equivalent in-phase and quadrature carrier method for FM, both paths send the same data resulting in a double sideband spectrum of a single carrier, which was regarded as redundant and therefore inefficient compared to the double-sideband spectra of the dual in-phase and quadrature carriers of QAM modems. Also, single-sideband (SSB) transmission was considered undesirable for modems because there is no simple way to efficiently demodulate an AM-SSB signal without a carrier reference. FM was dismissed as an analog technique totally unrelated to data transfer except with regard to FSK. In an ordinary modem, binary data is normally passed through a baseband raised-cosine filter which limits the bandwidth of the baseband signal so that when one multiplies the baseband signal by a carrier, control over the passband signal bandwidth is provided without intersymbol interference. The output of an ordinary modem includes signals having discrete phases such that data included therein can be identified by discerning the phase of each bit. For instance, whenever a signal has a +90° phase shift it is interpreted as 0 and when the phase shift is −90° it represents 1, etc. Thus, in an ordinary modem employing a carrier, the phase and/or amplitude of the carrier signal are determined by the current symbol being transmitted. The carrier assumes only selected values of phase and amplitude for most of the duration of each symbol and graphical plots of all the selected phase-amplitude pairs are called the constellation points of the modem. ordinary modems require that there be distinct points in the constellation for each possible value of the transmitted symbol. Furthermore, bit errors occur in the receiver if the points are mis-assigned due to intersymbol interference or due to noise on the link. In many applications the computational requirements of the modem introduce a delay which is detrimental to the operation of the system. For example, digital voice transmission and multiple access networks are sensitive to delay in the modem. Furthermore, the rate at which a modem may transmit and receive data per unit of bandwidth is called the modem bandwidth efficiency. In the discipline of Digital Information Theory this efficiency is known to be maximized when the transmitted signal has the maximum entropy or randomness. The maximum entropy transmission is band-limited Guassian noise, and among other properties Guassian noise will not dwell at a distinct phase-amplitude pair as in a constellation. Thus it is desirable to provide a modem having a passband carrier which minimizes the internal processing time and also maximizes the bandwidth efficiency without sacrificing bit error rate. In accordance with the present invention, a method of communicating a sequential series of symbols over a transmission link comprises the steps of multi-rate polyphase filtering the symbols, using the filtered output to modulate a carrier, transmitting the modulated carrier across the transmission link, receiving the modulated signal, applying the inverse of the transmitter polyphase filter to the received signal, and thresholding and re-assembling the output of the inverse filter to recover the transmitted symbols. Data is transmitted from a first modem to a second modem across a wireless transmission link by forming input data frames from the input data, multiplying the input frames by a rotation matrix, frequency modulating and transmitting the rotation matrix output, receiving and frequency demodulating the transmitted data, multiplying the demodulated signal by a second rotation matrix, and re-assembling the de-rotated data to recover the original data. The presently disclosed modem for communicating symbols across a transmission link includes a transmit portion and a receive portion, wherein the transmit portion comprises a partitioning element for dividing the input into parallel data channels, a baseband transmit rotation section for polyphase transforming the channeled data into parallel signal channels, a re-arrangement of the parallel signal channels into sequential serial samples, a carrier modulator for providing a modulated signal, and a transmitter for transmitting the modulated signal. The receive portion comprises a receiver for receiving the transmitted, modulated signal, a demodulator for demodulating the received signal into parallel signal channels, a receive rotation section for polyphase transforming the demodulated, received signal into parallel data channels, and an assembling element for combining the parallel data channels into a serial data signal. The polyphase filtration of these methods and this modem enables FM modulation of the transmitted symbols since only a real component of the original symbols is generated. FM, whether achieved using frequency modulation or phase modulation of the input signal, provides enhanced immunity to non-Guassian noise, provides high bandwidth efficiency, utilizes non-coherent IF and thus requires no carrier recovery, is of lower cost than conventional modems in part due to the absence of A/D converters, provides low co-interference due to the FM capture effect, and is compatible with analog signals, owing to the use of commuting operators. More power efficient, but potentially less linear amplifiers such as Class B and Class C can be employed since zero crossings are used to determine data content; carrier recovery is not required. Prior art modems employing modulation schemes such as QAM cannot employ such non-linear amplification. Further, satellite modems employing the presently disclosed method and modem save on TWT backoff power and thus are more energy efficient since intermodulation is not a problem, while personal computers achieve higher data rates without sacrificing bit error rate. The polyphase filtration of the foregoing embodiments is implemented in a first embodiment by a wavelet filter (e.g. Quadrature Mirror Filter) pair. The partitioning element partitions the serial data among plural, parallel data channels prior to linear-phase FIR vector filtration, the filter coefficients being square matrices, whereby the input data are transformed into parallel signal channels. The transformation is by way of a convolutional rotation of the input data vector. Each coordinate of the output signal is confined to a frequency sub-band which slightly overlaps its neighbor. Pre-emphasis in the transmit portion, prior to the rotation in a first embodiment, places most of the information in the lower baseband frequencies. This is due to the noise probability density function of an FM discriminator that is proportional to the square of the frequency. De-emphasis in the receiver results in an addition to the overall gain equation. This equation, in one embodiment, includes contributions by the FM transmitter gain, the de-emphasis gain, and a noise reduction gain. The pulse amplitude levels representing the partitioned data bits within each sub-band need not necessarily correspond to an integer number of bits, as long as all of the levels in all the subbands correspond to an integer value. The receiver portion provides the de-rotation filter for performing a reverse transformation to recover the original data. In a first embodiment, the reverse transformation commutes with the modulation transformation. In a further embodiment, the coefficients of the de-rotation filter in the receiver are adaptively selected for equalization to correct for transmission path distortion, since the analyzer is a fractional-rate FIR filter. Thus, a near-perfect reconstruction filter is employed. A threshold operator takes the nearest integer coordinate values as the most likely symbol. In a further embodiment of the present invention, the commuting rotation and de-rotation filters are derived from the elementary matrices which describe a geometric rotation of a vector. Their function is to transform an input data vector within a data coordinate system into a signal vector within a signal coordinate system such that the sequentially serialized coordinates of the signal vector would form the digital samples of a bandlimited analog signal. Yet another approach to the matrices is mathematical transformation by way of discrete wavelet transformations. In each of the transmitter portion and receiver portion, the rotation operator is ideally a computationally efficient multi-rate wavelet filter bank. Logarithmic amplification of the baseband signal prior to introduction into the FM transmitter modulator results in an improvement in the modulation gain out of the receiver. Further, as a by-product of the logarithmic amplification prior to the transmitter and de-amplification after the receiver, noise introduced in the transmit channel is attenuated. FM double-side band is employed in systems requiring minimal cost and complexity. FM single-side band is easily implemented in systems in which the bandwidth reduction is desired. AM modulation of a carrier is also implemented in a further embodiment. The foregoing features of this invention as well as the invention itself may be more fully understood from the following detailed description of the drawings in which: Referring now to An analog input signal X fed to the analyzer input port 14 a is decomposed into a pair of signals W′ and V′ each of which is fed to a corresponding one of the output ports 16 a, 16 b. Similarly, a pair of input signals W′, V′ fed to the synthesizer input ports 18 a, 18 b are reconstructed in to an output signal Y at a synthesizer output port 22 a. The decomposition and reconstruction processes performed by the analyzer 12 and the synthesizer 20 as well as the signals V′ and W′ will be described further hereinbelow. Suffice it here to say that the input signal X is decomposed into the signals V′ and W′ such that the signals V′ and W′ may later be combined to exactly reconstruct the input signal X. It should be noted that the sub-analyzer and sub-synthesizer building blocks operate digitally, however, for clarity in the description, the signal conditioning circuits required to convert between analog signals and digital signals in the system have been omitted. Such signal conditioning circuits will be described in conjunction with The sub-synthesizer 20 performs the inverse operation of the sub-analyzer 12, and vice versa. That is, if the signals W′ and V′ from a sub-analyzer are applied as input signals to the sub-synthesizer input ports 18 a, 18 b, the output signal Y will be identical to the sequence of input samples, X, except for a predetermined time delay. In a preferred embodiment, the time delay corresponds to one sample time. Likewise, applying the signals V′, W′ to the respective synthesizer input ports 18 a, 18 b and then applying the resultant output signal Y to the analyzer input port 14 a should provide at the respective analyzer output ports 16 a, 16 b the original input signals V′ and W′. As described above, the analyzer 12 and synthesizer 20 provide signal decomposition and reconstruction functions. As will be described below, the sub-analyzer and sub-synthesizer may be used as building blocks and may be coupled to provide more complex circuits which may themselves be coupled to provide a variety of signal transmitting and receiving systems. Furthermore, the analyzer 12 and synthesizer 20 as well as other system building blocks to be described hereinbelow may be efficiently implemented in hardware, software or by a combination of hardware and software. The analyzers and synthesizers to be described herein operate on an ordered sequence of numbers, which may be, but are not limited to, samples from an analog to digital converter (ADC). For example, samples may be expressed as X(0), X(1), X(2), and X(3) where X(0) is the most recent sample. Each of the binary numbers X(0) through X(3) has a particular value within a predetermined range of values. An 8-bit ADC, for example, may provide a range of decimal values between −128 to +127. The sequence of samples X(0) through X(3) may be considered as the coordinates of a vector [X] in a 4-dimensional space. A linear transformation may be made from the coordinate system of X to another coordinate system in the same 4-dimensional space. Thus, the vector X may be transformed to a vector Y by a “rotation” matrix C. In matrix notation this may be expressed as:
Using axes of the new coordinate system that are mutually orthogonal, the vector Y has components which correspond to the projections of X onto the new transformation axes. Such projections are provided by forming a vector dot product. In a 4 dimension case for example, a set of Walsh codes, Code 1-Code 4, may be provided as:
The Walsh codes Code 1-Code 4 above represent an orthogonal coordinate axis in the 4-dimensions of this time ordered space. Codes 1-4 have a length corresponding to the square root of 4, (i.e. the dot product of code 1 with itself equals 4) and thus are not unit vectors. The rotation matrix C may be expressed as:
That is, the rows C_{1}-C_{4 }of the rotation matrix C correspond to the components of the Walsh code vectors such that matrix multiplication between the rotation matrix C and the vector X is equivalent to the dot product of a particular one of the vectors C_{1}-C_{4 }with the vector X. The components of the row vector X may be expressed as:
The linear rotation operation is exactly invertible, and thus:
In a two dimensional case, Walsh vectors may be expressed as C_{1}=[+1,+1] and C_{2}=[+1,−1]. Thus, analyzers may be defined having 2 outputs using the 2-dimensional vector above. Walsh vectors of any dimension may be generated by the two dimensional vectors C_{1}, C_{2}. That is, substitution of a set of Walsh codes into a 2-dimensional generator matrix provides a new Walsh code having twice the dimension. Using this procedure an N-dimensional transformation needed to provide an analyzer, may be provided. Thus, because of the way that Walsh vectors are constructed from their two dimensional Walsh generators, higher dimension analyzers and synthesizers may be constructed from the 2-dimensional case. With this matrix transformation method, matrix equations for providing a modem or a signal encryptor may be generated. The fundamental operation in the requisite computations is addition and subtraction of terms. It should be noted that in the case of a cascade analyzer or synthesizer provided using the matrix method with a matrix having N dimensions, the lowest frequency channel is operated on by the matrix vector C_{1 }and each subsequent channel i is operated on by a corresponding matrix vector C_{i }until the highest frequency channel is operated on by the matrix vector C_{N}. Thus, in the case of a 4 dimension matrix the lowest frequency channel is operated on by the matrix vector C_{1 }and the highest frequency channel is operated on by the matrix vector C_{4}. The case of a tree analyzer or synthesizer provided by the matrix method will be described in conjunction with In another preferred method the equations which describe the sub-analyzer output signals V′ and W′ may be provided as:
As will be explained further below the scaling function V′ and the residual W′ defined by Equations 1 and 2 may be evaluated on alternate input samples of the input signal X, such that the value of the residual W′ corresponds to twice the difference between the “center” sample, X(n−1), and the midpoint of a line connecting the two nearest neighbors of X(n−1) here, the two nearest neighbors being represented as X(n) and X(n−2). When Equations 1 and 2 are evaluated on alternate input samples, only the odd numbered (or alternatively only the even numbered) “center points” are selected for the calculation. Thus, the output rate of the sub-analyzer corresponds to one-half the input rate of the input signal X. The residual W′ may also be interpreted as a deceleration about the center point. Thus it is possible to define the residual W′ as an acceleration and substitute −W′ for W′ in the Equations 1 and 2 above to provide an alternate and essentially equivalent expression for the residual W′. Other implementations using more nearest neighbor points and which define the residual W′ as a higher order or first order derivative of the function evaluated about the center point, may also be used. The interpretation of the residual W′(N) described above indicates that the residual W′(N) responds only to changes in the slope around alternate samples. Thus one characteristic of the residual W′ in the present invention is that the residual W′ is provided having a value of zero when the slope about the center point is constant. The conventionally defined residual W(n) is related to the effective residual W′(n) as defined in the present invention by
As a consequence of the definition for the residual W′(n) in the present invention, signal compression based on linear or non-linear quantization of the residual W′(n) (rather than the residual W(n) as defined in the conventional approach) may result in improved performance over compression schemes based upon conventional definitions of the residual W(N) such as:
The sub-synthesizer operation may be described by equations 3 and 4 below:
By defining the variable SHIFT, as above and taking advantage of the fact that multiplication and division by a power of two is equivalent to a right or left shift on a binary computer, practical implementations of these equations may be provided with relatively little and simple hardware and are therefore preferred. Thus in the approach of the present invention, the residual W′ has been defined to provide a particular characteristic and the sequence V′ is selected to provide the remainder of the sequence. Referring now to For example, in a communication system it may be desirable to couple a plurality of say N sub-analyzers as described above where N is selected such that a signal V′_{N }is selected such that the signal V′_{N }is provided having a sample rate which is below twice the lower frequency cutoff of the communications link. Since each sub-analyzer halves the sample rate to its input, the various output signals, W′_{1}, W′_{2 }. . . W′_{N }of the cascade analyzer are provided at differing rates. Referring now to Referring now to The tree analyzer 28 may be provided by analyzing each of the residual sequence outputs W′ until all such sequences are brought down to one rate, which is normally below the Nyquist rate of the lower cutoff frequency of a transmission link with which the analyzer cooperates. The three level analyzer tree 28 is here provided having all of its output 29 a-29 h at ⅛ the sample rate, r, of the analog input signal, X. Each output sample may include many more bits per sample than the input sample, however, the numbers represented by those output samples will usually be small in magnitude and easily re-quantized by a quantizer (not shown). Referring now to It should be noted that in each of the applications to be described herein below in conjunction with In the case where analyzers and synthesizers are provided as tree type analyzers and synthesizers the matrix vectors C_{i }where i=1, 3, 5, etc. . . . operate on the channels in the bottom portions of the tree analyzer or synthesizer while the matrix vectors C_{i }i=2, 4, 6, etc . . . operate on channels in the top half of the tree analyzer or synthesizer. Thus, if the analyzer 28 having channels 29 a-29 h were provided by an 8 dimensional matrix having vectors C_{1 }through C_{8 }then channels 29 h through 29 e would be operated on, respectively, by matrix vectors C_{1}, C_{3}, C_{5}, and C_{7 }and channels 29 d through 29 a would be operated on, respectively, by matrix vectors C_{2}, C_{4}, C_{6}, and C_{8}. Referring now to It should be noted that here five sub-analyzers and sub-synthesizers are coupled as shown, however, those of ordinary skill in the art will now appreciate that any number of sub-analyzers and sub-synthesizers may be used. It should also be noted, and as described above, that although cascade analyzers and synthesizers are here shown, tree-type analyzers and synthesizers may also be used. An input signal X may be decomposed, therefore into any number N of signals. In general, and as mentioned above in conjunction with In general operational overview of the signal encryption circuit 32, the input signal X, which may be for example a voice signal, is fed to the analyzer input port 33 a′ and decomposed by the analyzer 33 in the Manner described above in conjunction with As shown here one method for encrypting the residuals W′_{1}-W′_{5 }is to provide a logical exclusive or (XOR) operation between the sign of the residuals W′_{1}-W′_{5 }and a logical variable having a value corresponding to a logical one (as shown). Alternatively, another method of encryption may be provided by adding several individual secret binary bit streams to one or all of the residuals W′. The residual signals W′ are generally provided having a relatively low power level and thus the residual signals W′_{1}-W′_{5 }may appear simply as noise signals (i.e. “buried” in noise) by either or both of the above mentioned encryption techniques. Other methods of encrypting well known to those of ordinary skill in the art may also be used including permutation of channels and substitution of symbols. It should also be noted that alternatively each of the signals V_{1}′-V_{5}′ may be encrypted or a combination of the signals W_{K}′ and V_{K}′ may be encrypted. It should also be noted that the signal encryptor 32 may be modified to include digital cipher feedback, which is common in the encryption art. To provide digital cipher feedback in the present invention, each of the 5 XOR outputs from the analyzer should be fed back to the inputs of 5 separate secret random number generators (not shown) in the manner of the data encryption standard (DES) for encryption. It should also be noted that any means well known to those of ordinary skill in the art for providing random digital numbers including but not limited to the DES, with or without cipher feedback may be used. Furthermore, other methods of deriving and using the random numbers to modify the value of the residuals W′ including but not limited to additive noise techniques and the like which do not use the XOR function may also be used. The encrypted signals E_{1}-E_{5 }are subsequently coupled to corresponding ones of the analyzer output ports 33 a-33 f and then fed to the synthesizer input ports 36 a-36 e along with an unmodified signal V5 fed to input port 36 f. The synthesizer 36 employs Equations 3 and 4 to reconstruct the signals E_{1}-E_{5 }fed thereto to provide a reconstructed output signal Y which may be transmitted on a communication channel (not shown) for example. A receiver (not shown) receives the transmitted signal and analyzes the signal to recover the signals V5 and E_{1}-E_{5}. The receiver subsequently decrypts the signals E_{1}-E_{5 }to recover the signals W_{1}-W_{5 }and re-synthesizes to get the original signal X. The synthesizer 36 reconstructs the signals E_{1}-E_{5 }such that the resultant reconstructed signal X′ provided at the synthesizer output port should not occupy more than the Nyquist bandwidth. Thus it is possible to have the sampling frequency of the highest frequency encrypted stage correspond to twice the upper cutoff frequency of the transmission channel. The reconstructed encrypted signal X′ provided at the synthesizer output port will of course have a different (noise-like) power spectrum and a different average total power due to the additive noise introduced. The input signal power may be diminished to restrict the total channel power. Means for restricting the total channel power may include means for transmitting the power factor to the receiver by using the amplitude of pilot tones which may be used to synchronize the decryptor and an ADC. These pilot tones may be used to derive the ADC clock by a phase locked loop for example and may be provided as explicit narrow band tones or, alternately the pilot tones may be provided as a secret coded sequence added to the residual W′ to hide the true value of the residual. The receiver would correlate the appropriate residual in order to establish time synchronization. Specific coded sequences (such as the GOLD codes and JPL ranging codes) having suitable correlation properties required for acquiring synchronization by this method are known in the art of spread spectrum communications. Furthermore the encryptor recovers the signal out of the receiver substantially exactly like the original input signal except for the presence of noise from the channel and noise-like fluctuations due to imprecise timing synchronization of the ADC converters in the transmitter and receiver ends. The signal encryptor may also include a means for line equalization (not shown). Such means are well known to those of ordinary skill in the art of modem design. Variations in line loss with frequency (tilt) and phase shift may be compensated by adaptive filtering and often includes a precursor burst of known energy to set the filter parameters and a means for periodically modifying the equalizer coefficients based on a measure of the quality of the received signal. It should also be noted that if in the signal encryptor 32 the value of the input signal X is set to zero, then the signal encryptor output signal X′ would correspond to random noise generated by the encrypting random number generator. Thus in this case the signal encryptor 32 acts as a broadband random number generator. Thus, the signal encryptor may be used as a generator to provide a signal on the line which may be used in the line equalization process. Furthermore the analyzer may be used to measure the channel response of each channel. If the synthesizer, thus provides a test signal on each channel the analyzer may measure the response on each channel to determine the loss on each channel. Thus the analyzer and synthesizer may be used to provide a method of line equalization. As will be described below in conjunction with The delay in a tree may correspond to
Referring now to The analog signal fed to the ADC 43 input port may be fed through an amplifier or other preconditioning circuits (not shown) and then fed to the ADC 43. Preferably signal preconditioning circuits such as low noise amplifiers and buffer amplifiers are relatively wide-band amplifiers and are further characterized as having relatively low levels of phase dispersion over the bandwidth of the amplifier. That is, the amplifiers impart to the amplified output signal a substantially equal phase shift to the amplified output signals therefrom at least over the bandwidth of the transmitted signal. Further, the sampling rate of the ADC 43 is preferably greater than twice the Nyquist sampling frequency (i.e. greater than twice the frequency of the highest frequency component signal in the input spectrum). The ADC 43 converts the signals provided from the filter 42 in accordance with the predetermined sampling rate to provide a stream of digital words. At the output of the ADC 43 such stream of digital words are fed to a signal encryption circuit 44. The signal encryption circuit 44 includes an analyzer 46 which appropriately decomposes the signal fed thereto into a plurality of signals, an encryptor circuit 48 which encrypts the decomposed signals and a synthesizer 50 which combines the signals fed thereto into a reconstructed encrypted signal. The signal encryption circuit 44 may be of the type and operate in a manner similar to the signal encryption circuit 32 described above in conjunction with The encryptor circuit 44 may encrypt the signal by adding a secret number to the residue, W′ with the value of the secret number known only to the sender and receiver. Such an addition may typically be done modula 2. That is, by providing a logical exclusive-or (XOR) function bit by bit to the residual signal. Such addition may also be accomplished by simply adding modulo the actual word size of the data. The encryptor 48 may be provided as a secret number generator which may be of a type that provides a different secret number based on the number of times it is requested to do so, or it may be of the type that the output depends only on its input. The latter type may often be self-synchronizing. The former type may suffer loss of sync between the encryptor and the decryptor if the receiver, for any reason, loses bit or word synchronization with the transmitter. The encryptor 48 may provide permutation operations among the several channels or alternatively the encryptor 48 may provide substitution operations or alternatively still the signal encryptor may provide the logical or operation described above in conjunction with The signal encryption circuit 44 feeds the reconstructed encrypted signal to an input port of an output signal conditioning circuit 51 which includes a digital-to analog-converter (DAC) 52 which may be provided as a logarithmic ADC for example, which converts the encrypted bit stream fed thereto to an analog signal representative of the encrypted bit stream. The analog signal is subsequently fed to an output filter 54 having appropriately selected filter characteristics as described above. The filter 54 couples the signal fed thereto to a first end of a transmission channel 56, which may be provided for example as a telephone line. A second end of the transmission channel 56 is coupled to the receiving portion 40 b of the transmitting and receiving system 40. In general overview, the receiving portion 40 b receives a secure signal fed thereto and decrypts the signal to provide a clear text signal at an output port. The receiving portion 40 b of the transmitting and receiving system 40 receives the encrypted signal through an input signal conditioning circuit 57 at an input port of an input filter 58 having appropriately selected filter characteristics. The input filter 58 couples the signal to an ADC 60 which converts the analog signal to provide a stream of digital words in the same manner as described above. The filter 58 may be provided having a low pass filter frequency cutoff characteristic corresponding to one-half the sampling frequency of the ADC 60. The ADC 60 feeds the stream of digital words to a decryption circuit 61 which includes an analyzer 62 which decomposes the signal fed thereto into a plurality of signals and a decryptor circuit 64. If the encryptor circuit 48 in the transmitting portion 40 a of the system used a secret number to encrypt the signal, the decryptor circuit 64 subtracts the secret number, to recover the original data. The decryptor-synthesizer combination 64, 66 performs the inverse operation of the encryptor-analyzer 48, 46 at the transmitter and feeds the decrypted and appropriately recomposed signal to a digital to analog converter (DAC) 68. The DAC 68 provides an analog signal corresponding to the bit stream fed thereto and feeds the analog signal to a receiver output filter having appropriately selected filter characteristics as described above. If the analyzer and synthesizer are provided as the cascade type or the tree type, then as mentioned above, the number of levels may typically be determined by the lower limit of the input signal bandwidth. Thus a practical requirement for most applications may be to dispose a pre-filter (not shown) having a high-pass or band pass filter characteristic prior to the signal conditioning circuit 41. In any case, whether such a filter is or is not provided, the system of sender and receiver may not recover a signal having a bandwidth which exceeds the bandwidth of the communications link 56. In applications where the communication link 56 is provided by electromagnetic or sonic energy, the bandwidth of the link generally has practical limits, for example a single television (TV) channel. Referring now to The encrypted signal is subsequently fed from the signal encryptor 78 to an input port of a DAC 82 which converts the encrypted bit stream to an analog signal. The analog signal is coupled through a filter 84 which provides an appropriately filtered signal to a first input port of a signal combining circuit 86. A timing circuit 87 includes a timing signal generator 88 for providing a timing signal. The timing circuit 87 may provide to a receiver 92 either an analog or a digital timing signal and thus an optional ADC 90 is here shown coupled between the timing signal generator 88 and the summing circuit 86. If the DAC 90 is included in the timing circuit 90 then the timing signal is fed through optional signal path 91′ to the digital input port of the DAC 8 d. If the ADC 90 is omitted then the timing signal is fed through signal path 91 to the summing circuit 86 as shown. It should be noted and those of ordinary skill in the art will recognize that implicit in the recovery of information at the receiving end of an analog communications link using the analyzer and synthesizer building blocks operating according to the above equations is the need to synchronize the ADC in the transmitting portion of the system and the DAC at the receiving portion of the system such that the transmit and receive portions of the system agree on the exact time of sampling. Large inaccuracies in synchronization will result in gibberish out of the receiver. Small statistical jitter in the timing synchronization will have the same effect as noise on the link. Timing and synchronization schemes are known to those of ordinary skill in the art. In one method for example, an oscillator locked to (i.e. derived from) the sampling rate of the sending portion may be sent over the link. The received oscillator signal may be used to derive (via phase locked loop techniques for example) the receiver's clock signals for sampling. As mentioned in the secure transmitting and receiving system 40 ( In those systems employing cascade-type or tree-type analyzers and synthesizers ( One method for removing an interfering signal (namely, the received oscillator signal) from the accompanying system information signal is to form a feedback loop that subtracts out an amount of the known (as received) frequency until the resulting difference no longer contains any narrow band correlation to the known (received) oscillator. The operation of a phase locked loop includes multiplication of an input signal with a local oscillator (which, in turn, is locked to the received timing signal) and then integrating the result with a low-pass filter. Another method for timing synchronization between the sending and receiving portions of the system will be described below in conjunction with Referring now to The synthesizer 98 feeds the bit stream into an input port of a DAC 100, which may be provided having a nonlinear response characteristic, which generates an analog signal corresponding to the bit stream fed thereto. In accordance with the Nyquist sampling theorem procedure, the digitized samples out of the sender should be converted via the DAC 100 to an analog signal representation of the digital signal fed thereto. The analog signal is subsequently filtered for anti-aliasing with a filter 102 having a low pass filter characteristic and preferably having a relatively steep filter skirt and a cutoff frequency corresponding to one-half the sampling rate frequency. The filtered signal is coupled to a first input port of a summing circuit 104. A timing circuit 106 feeds a timing signal into a second input port of the summing circuit. The summing circuit thus superimposes the two analog signals fed thereto. Alternatively, a timing signal may be transmitted via the input to the synthesizer 98. The superimposed analog signal is transmitted over an analog transmission link 107 (e.g. a telephone line) to a receiver 108 where the timing signal may be used to provide timing data for the receiver such that the bit stream may be recovered from the analog signal. It is believed that a modem constructed in accordance with the present invention may operate at or near the maximum data rate theoretically possible for a transmission link based on signal to noise ratio of the link and Shannon's Law. At the receiver 108, substantially all noise in the frequency range above one-half the receiver sampling rate frequency should be filtered with a filter having a low-pass filter characteristic and having a cutoff frequency above one-half the receiver sampling rate frequency. If the transmission link 107 is provided as a transmission line over which signals having a frequency between 400 Hz and 3200 Hz may be transmitted then the input signal may appropriately be sampled at a sampling rate typically of about 6400 bps. A single tone having a frequency typically of about 3200 Hz may be provided by the timing circuit and added to the transmitted signal and phase-locked at the receiver as one means of synchronizing a receiver ADC clock. Since the tone may be provided having a known amplitude and frequency, it may be subtracted rather than filtered out at the receiver and thus there is no resultant loss of data. Likewise a signal tone having a frequency typically of about 400 Hz may be used to provide a word synchronization for formulating the word comprising the input to all the cascade levels. Signals in the frequency range below 400 Hz may be used for signalling for line turnaround in a semi-full duplex modem, for sending reverse channel data and network information. Bits per cascade level may be increased until the maximum power per unit of data which may be transmitted and Shannon's limit are reached. Furthermore error correcting codes such as M of N codes (M redundant bits out of N) and scrambling signals may be applied to the input data words as is known to those of ordinary skill in the modem art. Referring now to The coded modem 110 includes a coder 112 for providing the coding operation. The coded signal is fed to a synthesizer 114 which provides a relatively broadband signal having a noise-like frequency spectrum to an input port of a summing circuit 116. A sharing signal may be fed to a second input port of the summing circuit. The summing circuit couples the signals fed thereto to a first end of a transmission line 117. At the demodulator portion, a signal tap 118 couples a portion of the signal transmitted over the transmission line 117 to an optional signal processor 124. The signal processor 124 feeds a processed signal to a receiver 125. The codes of the coder 112 are selected to have good auto- and cross-correlation properties, and thus the modem data may be recovered even when the modem operates at low power relative to the shared signal. To a sharing device, the receiver 125 which may be provided as a TV receiver for example, the modem signal may appear to be a small random background noise. However if the shared signal is coupled to the TV receiver 125 through the signal processor 124 containing the code sequences, C, then most or substantially all of the modem “interference” in the TV receiver 124 may be removed by known techniques for cancelling correlated noise. The modem 110 includes the coder 112 for providing a coding operation C, and a decoder 122 for providing a correlation operation, C{circumflex over ( )}. In a correlation operation, data is recovered by digitally integrating the product of the received sequence with the stored code, C. A signal sharing the link will typically tend to integrate to near zero since the shared transmission is uncorrelated to the chosen code, C. The optional interference cancellation operation for the sharing signal is performed by the signal processor, 124. A method for timing synchronization between the sending portion and receiving portion of the system 110, is now described, however before describing the method it should be noted that although the method may be more clearly explained within the context of the coded modem, the method may also be applied, with minor variations, to other systems including the signal encryption systems described above in conjunction with At least one of the signals, W′_{k}, is forced to be a sequence having known correlation properties. In a cascade synthesizer, since “word” synchronization is required, the chosen signal W′_{k }may preferably correspond to the signal having lowest inband sample rate. It should be noted however, that in a direct sequence coded modem all the W′_{k}s are so coded. There are many examples of suitable code sequences such as JPL, GOLD codes and Walsh codes. For illustration and not by way of limitation, the Walsh codes (also known as Hadamard codes) are described. The kernel for a Walsh code is provided as:
Higher order codes are found by substitution of a level into the kernel as shown
Many other codes, (such as the GOLD codes) are known that have “good” correlation properties. By correlation is meant multiplication and integration which in a two valued binary case (+1,−1) reduces to just the vector dot product of the sequences. A dot product between two identical codes provides a predetermined output (i.e. (Code 1) DOT (Code 1)=4) However, a dot product between 2 unlike codes would provide an output of zero (i.e. (code 1) DOT (codes 2,3,4,)=0) Similarly, this relationship also holds true for each of the other 3 codes. Thus, these are orthogonal codes. Non-orthogonal codes having a large auto-correlation and a small cross-correlation may also be suitable, and some of such codes are known to be particularly good for fast acquisition of synchronization in a sliding correlator. An example of a sliding correlator may be made with code 3. If word sync is unknown (assume for this discussion that bit sync is known) then one of four possibilities may occur in the receiver's correlator: they are
Correlation, that is the dot product of the receiver's code with each of the 4 possible patterns reveals that the correlator will compute a dot product of 0, −4, 0, and +4 for each of the 4 possibilities, respectively. However, only the correct word synchronization (i.e. namely possibility 4) will have large positive (i.e. +4) correlation. By sliding the receiver's assumed clock, bit by bit relative to the incoming signal, and performing the correlation, the receiver may thus find word synchronization, hence the name “sliding correlation”. Thus, it should also be noted that the maximum correlation will occur when both the word and bit synchronization are correct. To accomplish the foregoing method of synchronization, the residual, W′, of the lowest frequency stage in the sender's cascade may be arranged to have code 3 define its value (or at least the sign of W′ would follow code 3). The preceding discussion illustrates some of the many ways receiver synchronization may be achieved. In some system applications such as encryption, the timing from the receiver may also drive other building blocks, such as a decryptor. In full duplex operation, a clock in the receiver may also be used for transmission from that end such that there is only one master clocking the system. Assuming that bit and word synchronization have been established in the modem 110, then an incoming data bit may be expressed as corresponding to either +1 or −1. If that data bit multiplies one of the codes, code 2 for example, then the resulting 4 bit sequence is either code 2 or code 2 with the sign of each bit reversed. If that sequence is applied to one of the W′ inputs of a cascade or tree synthesizer as a sequence of bits, then the receiver's analyzer will recover that W′ and may correlate with code 2 to get either a large positive or a large negative number which will determine the receiver output as a +1 or a −1 respectively. Here for ease of explanation, an example using a single bit has been described. In practical systems of course, such operations would typically be performed on digital words having a plurality of bits. The application of the coded data bits to the W′ inputs of a cascade synthesizer is somewhat complicated by the fact that each stage in the cascade operates at a different sampling rate. Such an operation may be more easily accomplished in a tree synthesizer since input data may be assembled into words and applied all at once at the frequency of the lowest synthesizer stage. For a tree synthesizer, the transmitter's power would also be distributed more evenly across the link bandwidth—a preferred and efficient case. Spreading the transmitter's energy evenly across the link bandwidth is a prescription for possibly achieving operation at the maximum limit of the link. Several coded modems of the type described in A two wire full duplex modem may be provided by using a companion set of almost orthogonal codes. The Walsh codes identified as code 1 through code 4 above, and their bit-wise complement, are only one half of the 16 possible combinations of 4-bits. As shown below the remaining combinations also form another set of mutually orthogonal vectors, below numbered as c5 through c8. This second set of 4 vectors is not orthogonal to the first set. It can be described as “nearly orthogonal” because the dot product of any member of set 1 with any member of set 2 is always half the length of the vector; and of course the dot product of any member with another of the same set is always 0 except that the product with itself is always equal to the length.
The master group is orthogonal and the slave group is also orthogonal, however the cross group correlation is −2 for the dual and +2 for any other cross term. One end of a transmission link may transmit signals using the master set of codes and a second end of the transmission link may transmit using the slave set of codes. It should be noted that the same benefits may be realized by using the matrix transformation approach to provide the synthesizer 114 and analyzer 120. Thus to provide full duplex operation, the modems at each end of a link can be assigned to use set 1 or set 2 as Master and Slave. If the Master used only codes 1 and 2 and the Slave used codes 3 and 4 then all echo signals would be totally cancelled by the orthogonality but the data throughout for each modem would be half of the rate possible by using the arrangement described above based on the two sets of codes. Furthermore, coded modems of the type described in conjunction with For a modem 126 as shown in One application for such a coded modem would be to take advantage of the processing gain of the correlation receiver to recover a low power signal from the modem buried in a large “jamming” signal. Practical examples of jamming signals include voice (hence data may be sent as ‘noise’ under the voice), television (hence high definition digital information may be sent in the same channel as standard video to maintain compatibility), code division multiplexing and two wire full duplex FDX. The disclosed technique is an improvement on those methods because the modem utilizes the bandwidth more efficiently than heretofore. Furthermore, the modem described herein, being a modulator (the synthesizer) and a demodulator (the analyzer) may also take the form of a baseband RF or soundwave transmitter modulator and receiver (or above baseband except for the limits of ADC's). Such a receiver may have application in receiving digital High Definition TV (HDTV). Coded modems of the type described above in conjunction with Referring now to In principal the coded tree modem 126 operates such that the coder circuit performs a coding operation of multiplying a data word times an orthogonal code C. The decoder performs a correlation operation here denoted C{circumflex over ( )}. It should be noted that the final V′ input sequence on line 130 h may be set to zero if it is assumed to be below a frequency corresponding to the lower frequency limit of the passband frequency of the link 132. Referring now to An output signal conditioning circuit 149 includes a DAC 150 coupled to the output port of the synthesizer 148. The DAC 150 receives a digital stream of words and provides an analog output signal representative of the bit stream fed thereto. A filter 152 having appropriately selected filter characteristics couples the analog signal from the DAC 150 to a first end of a transmission line 154. A second end of the transmission line 154 is coupled to a receive portion 138 b of the system 138. The receive portion b includes an input signal conditioning circuit 156, which appropriately filters and converts the analog signal fed thereto to provide an appropriate stream of digital words to the first input port of an analyzer 158. An inverse quantizer 160 (i.e requantizer) is coupled between the analyzer 158 and a synthesizer 162. In operation, the requantizer 160 remaps to the original bit definition. The compression operation, of course, lowers the information content of the signal and that lost information may not be recovered. In many applications, however, the lost information is redundant or the human observer is insensitive to the level of detail contained in the information discarded by the quantizer and thus little or no signal degradation may be detected. An output signal conditioning circuit 163 receives a reconstructed digital stream of words from the synthesizer 162 and provides an appropriately filtered analog output signal representative of the bit stream fed thereto to an the output port of the receive portion 138 b of the system. In speech compression processes, bandwidth may generally be reduced by limiting the number of bits devoted to the residual W′. However, an alternative which may provide an increased reduction in bandwidth may be accomplished by first replacing a signal V_{5 }on the cascade channel corresponding to the lowest frequency band with a zero, thus only transmitting the signal W's. Next, a signal W′_{3 }associated with a cascade channel corresponding to the frequency band in the 700 Hz to 1400 Hz frequency range may be eliminated or coarsely quantized. Furthermore, Huffman coding or codebook vector quantization methods may be used on W′_{2}. By adjusting the sample rate, the frequency band from 700 Hz to 1400 Hz may be isolated. Since human voice especially in the English language, generally does not include a formant in this range, the cascade channel corresponding to the frequency band from 700 Hz to 1400 Hz may be eliminated (i.e. by setting the residual W′_{3 }equal to zero) with little loss in intelligibility. Similarly, as shown in the Table below, W′_{1 }and W′_{5 }may also be set to zero.
Thus, in this example W′_{1 }through W′_{5 }and V_{5 }are sampled at the rates shown in the Table and it is possible to transmit only the signals corresponding to the residuals W′_{2 }and W′_{4}, for example, which have sample rates of 700 and 2800 baud perhaps less than 2 bits for each residual W′_{21 }W′_{4 }after Huffman coding. Further reduction may also be possible since W′_{2 }may simply be considered as another sampled signal and thus may also be subdivided by multi-resolution analysis to further reduce the bandwidth. By way of example, if 1.5 bits are used for the 700 sample/second residual and W′_{2 }is decomposed into 1400 samples/second, 700 samples/second, 350 samples/second, 175 samples/second, and 65 samples/second at 1.5 bits each then the total number of bits per second (bps) corresponds to 5085 bps to which should be added overhead bits for frame synchronization. This method is considerably less computationally complex than methods such as Linear Prediction Coding 10 (LPC10) and dynamic Excitation LPC and refinements thereto as are known to those or ordinary skill in the art of voice compression. Although not here shown, the receiver may be provided having the same form with received signal applied to V_{0 }and the clear output taken from V′_{0}. In view of the above, those of ordinary skill in the art will now recognize that combinations of the system described above may be created to form, for example, an encrypted data compression system for use on analog links. This may be particularly useful, for example, in those applications such as transmission and reception of high definition television signals in which the amount of digital data to be transmitted exceeds the Shannon Law limit of the link. Thus, in such applications the data may first be compressed by any algorithmic means until the sample rate is consistent with the Nyquist sampling theorem limit of the link and the data may then be applied to one of the modem systems described hereinabove in conjunction with Referring now to Referring now to The synthesizer 180 may be provided in accordance with the matrix transformation methods described above in conjunction with In the matrix approach, the synthesizer 180 operates on the data frames fed thereto with a rotation matrix. Here, an 8 dimensional rotation matrix would be applied to the data frames. Similarly, the analyzer 188 would perform an inverse rotation operation by applying an 8 dimensional matrix corresponding to the inverse matrix of the matrix used by the synthesizer 180. As mentioned above, the basic sampling rate of the DAC 182 and ADC 186 is 6400 samples/sec. The digital input and output operate at a frame rate of ⅛ or 800 frames per second. Each frame may be composed, for example, of 35 bits divided into 7 words of 5 bits (or more) each and applied to channels 180 a-180 h. The thruput is 28,000 bps (35 bits times 800 frames/second). The usable rate is determined by the S/N ratio on the link 184 and forward error correction (FEC). Thus, the modem 178 may operate at a data transfer rate at or below 28 kbps. Furthermore, the modem 178 may utilize FEC and as is common practice, the frequency range from 300 Hz to 400 Hz may be used for frequency shift keying (FSK) diagnostic signalling. Channel 180 h corresponds to signaling below 400 hz and cannot be used for data in this example. However, if a signal having a constant amplitude and alternating sign is applied to channel 180 h then a 400 hz tone may be filtered from the line signal by a receiver (not shown) to aid in synchronization. In addition, the known preselected amplitude may be recovered from channel 188 h as data and used as a gain calibration signal in the receiver and, may further be used to define block boundaries for FEC block coding schemes. Accurate timing synchronization and gain calibration are important to the operation of the modem 178. As described above in conjunction with The data in the channels 180 a-180 g may be scrambled, as is common practice in modems, so that the output appears more noise-like when transmitting the commonly encountered input string of 35 zeros or ones. Without scrambling, an input string of zeros will produce a modulated but strongly correlated output without a DC component. The data received by the analyzer 188 will be a multiple of the actual data. The multiplication factor will correspond to the dimension of the rotation matrix which in turn corresponds to the number of channels in the synthesizer 180. The receiver should quantize, that is, round the received channel output to the nearest multiple then divide by the multiple to reduce the effects of noise on the link 184. A modem using baseband modulation techniques, in which no carrier signal exists, may also be provided by using the matrix rotation approach or by the equations provided in conjunction with Thus, in a baseband system a modem operating according to the matrix method described above should modulate and demodulate groups of samples. For example, if the matrix vectors are of length eight, then eight samples should be processed together as an independent group by the demodulator. That is, the demodulator multiplies the group of samples by the inverse matrix used in the modulator. It should be noted that the grouping of samples is part of the modulation and distinct, for example, from a block data coding method for error correction, and both techniques may be simultaneously used in a single modem. In summary, a baseband modem technique uses an invertible mapping. According to the foregoing disclosure, and the issued parent patent, U.S. Pat. No. 5,367,516, the modulation and demodulation mappings may be characterized as a filter bank synthesizer and analyzer, or a rotation and counter-rotation matrix, or a baseband mathematical transformation and its inverse transformation. These characterizations may be distinct, or in some instances they may describe the same operations in different technology. The terminology and unification of the three characterizations appear in publications cited in conjunction with the filing of the instant patent application. For example, the analyzer and synthesizer in The invertible mapping for the modem is similar to a geometric rotation, although the mapping may or may not have a pure delay due to the filter. A sequence of samples X(0) through X(3) may be considered as the coordinates of a vector X in a 4-dimensional space. A linear transformation may be made from the coordinate system of X to another coordinate system in the same 4-dimensional space. Thus the vector X may be transformed to a vector Y by a “rotation” matrix, or by a filter bank, or by a mathematical transformation such as the DWT. The components of vector X can be equated to a frame of successive samples into the transmitter's D/A converter. The components of Y are assigned to those inputs to the filter bank corresponding to the sub-bands that are within the transmission bandwidth. In a modem based on the foregoing disclosure, and the issued parent patent, U.S. Pat. No. 5,367,516, X and Y are two different coordinate representations of the same vector. The modem modulator can be considered as a linear “rotation” operator [M] and the demodulator as an operator [D]. At the transmitter, it follows that X=[M]Y. The vector Y is being sent so the received data [D]X=Y if the modem is correctly conveying the data, a condition met by the requirement that [D][M]=z[I], where [I] is the identity matrix and z represents a pure, frequency independent delay, if any, through the system. If there is additive noise at the receiver then the linear demodulation operator must remove the noise term by a non-linear threshold operation as is common in all modems. In a general sense that includes both linear and non-linear operators, all modems require that [D][M]=z[I] in order to recover data correctly. However, in the baseband modem of the foregoing, and in the issued parent patent, U.S. Pat. No. 5,367,516, the geometric rotation analogy results because its linear operators commute. That is [D][M]=[M][D], or at least they can very nearly commute over the entire useful bandwidth of the channel. Geometric rotations in 2-dimensions commute and rotations in any dimension can be made to commute by making a series of 2-dimensional rotations that commute with a reversed sequence of counter-rotations, which is the essence of Vaidyanathan et al.'s design procedure for multi-dimensional (multi-band) QMF banks of the cited publications. As described in the foregoing, and in the issued parent patent, U.S. Pat. No. 5,367,516, any sampled analog signal can be framed as a vector A and can be digitally encrypted without bandwidth expansion and without digital compression since an encryptor at the transmitter can transmit X=[M][e][D]A and the receiving decryptor can compute A =[M][d][D]X, where digital encryption [e] and decryption [d] satisfy [d][e]=[I] and assuming [M] and [D] commute. Passband modems modulate one or more carriers with one or more separate baseband modem waveforms that represent data. The carrier modulation and subsequent demodulation is most often a linear operation in contemporary modems. Thus, at some carrier frequency, a modem may add a sine and cosine carrier wherein each carrier has been linearly amplitude modulated by baseband data modulators, or filters. The resultant signal may have both phase and/or amplitude variations as in a Quadrature Amplitude Modulation (QAM) modem. The baseband modulation of the foregoing disclosure can be applied in this same manner to make a linear modem for passband. Non-linear modulations into passband are also possible for data modems and analog signal encryptors. Non-linear modems have seldom been used except in Frequency Shift Keying (FSK) modems, which are very inefficient. However, non-linear modems employing the methods described herein may be more bandwidth efficient than any linear modulation including QAM when operated, for example, in the region of interest for wireless communications, assuming for comparison that there is no Forward Error Correction (FEC) for either the non-linear or linear designs. An exemplary non-linear FM modem is described herein. An FM DSB signal can be generated by the quadrature carrier method wherein a sine and cosine carrier are amplitude modulated, respectively, by the sine and cosine of the integral of the baseband modem signal. Thus, a cosine function amplitude modulates the cosine carrier, etc., so that the FM signal is indeed a non-linear modulation and distinct from linear passband modulations, which may or may not include baseband integrators. FM, and nonlinear Phase Modulation (PM), can also be generated by a voltage controlled oscillator (VCO) and by other techniques. The viewpoint taken in the foregoing, and in the issued parent patent, U.S. Pat. No. 5,367,516, that data and modulated signals can be different coordinate representations of the same vector, provides insight and specific techniques for the design of non-linear modems using rotations by filter banks or wavelets. In Typically, the output signal from the transmit portion 202 of one modem 200 will not be received by the receive portion 204 of the same modem 200. However, it is possible that a transmission path over which the output signal is transmitted and from which the input signal is received includes a memory device for storing the transmitted signal. In such case, it is possible that the same modem will carry out both the transmit and receive functions. For example, in alternative embodiments of the present invention, the memory device is a magnetic disk or a non-volatile solid-state memory device; modulated digital information to be stored in analog form is later retrieved and demodulated by the same modem. The memory device behaves as a modem transmission link with an exceptionally long link delay, so the bit-error-rate calculations and maximum system transfer rate into or out of the memory can be determined as for any modem system. In particular, the maximum error-free bit rate for recovery of the data is given by Shannon's law, where the signal-to-noise ration depends on the device physics as well as the noise figure of the electronics. On the receive portion 204 side of the modem 200, an input signal from an FM receiver (not shown) is received at a receiver interface 214, and is forwarded to an FM demodulator 216. The demodulated signal, after non-linear de-compression attenuation 217, is then counter-rotated by a receive rotation operator 218. After complimentary de-emphasis amplification 219, the result is assembled 220 into output data which is ideally the same as the original input data. The de-compression attenuation 217 further comprises an equalizer in an alternative embodiment. Modems employing convolutional rotation, such as discussed in the foregoing and in the parent patent, U.S. Pat. No. 5,367,516, are optimal in the sense of potential bandwidth efficiency. Encryption using this technique is optimal in the sense that the signal encryptor can transform any band limited analog signal into the digital domain, digitally encrypt it, and then transform the signal back to the analog domain without changing the analog bandwidth. This is in contrast to contemporary digital voice encryptors which rely on digital voice compression algorithms to achieve encryption without bandwidth expansion. Both of these optimal properties are a consequence of the invertible baseband transformation used in the modem disclosed in the foregoing and in the parent patent, U.S. Pat. No. 5,367,516. A preferred embodiment of this invertible baseband transformation is the Quadrature Mirror Filter (QMF) bank, also known as the wavelet filter, which can provide lossless reconstruction of signals over all of the filter's passband. Since the transition region to the stopband at the band edges can be designed to vanish as the filter delay increases, the modulator satisfies Shannon's criterion for no entropy loss in the mathematical limit. The rotation operator in a first embodiment is a multi-rate wavelet filter-bank. Such filters are designed in a manner similar to sub-band coding schemes. Each analyzer input corresponds to one of M overlapping sub-bands. Ports corresponding to out-of-band regions of the signal spectrum are not used for data, though they are used in a further embodiment to carry baseband synchronizing bits without violating spectral constraints. Regardless of the number of sub-bands, the polyphase rotation matrices commute exactly over the entire band. There is a very small entropy loss at the band edge (but not between sub-bands), as determined by the independently chosen filter length. For a time-frame corresponding to M samples, the input binary information is partitioned into the integer coordinates of an information vector. The output of the demodulation operator may be viewed as a vector with non-integer coordinates. “Thresholding” is a non-linear operation that removes noise by rounding the received vector coordinates into integers. It is normal modem practice when transmitting data to encode the data as equally spaced signed odd-integer values. For example, two bits are encoded as one of four levels, −3, −1, +1, +3. In this case, thresholding rounds to the nearest odd-integer value. The modem can also be used to transmit any bandlimited analog signal which has been digitally processed by sub-rate filtering, including but not limited to the digital processing step of digital encryption as described in the foregoing and in the parent patent, U.S. Pat. No. 5,367,516. When the modem is being used in this manner, the input samples to the modem modulator may be considered as signed integers spanning even and odd values including zero. Receiver thresholding rounds to those values. To be consistent in notation, integers are assumed to be input to the modulator, the modulator sends integers to the D/A converter and the receiver obtains integers from the A/D. The demodulator outputs integers which are thresholded by appropriately rounding the quotient to integers or odd-integers after dividing by the rotation matrix normalization constant. As described in the foregoing and in the parent patent, U.S. Pat. No. 5,367,516, this rotation gain is computed from the rotation operators (the filter bank polyphase matrix). The transmitted data, including synchronizing information, is represented by an M-dimensional vector. In one embodiment, the modulo addition of a secret vector to the carrying vector in the data representation results in a highly secure, digitally encrypted, analog signal out of the D/A. Decryption modulo-subtracts the secret vector to recover the carrying vector. Input information bits can be assigned to the data-representation coordinates of a vector and multiplied at the chip-rate by a pseudo-random sequence unique to each coordinate. The chip-rate result is then input to the modem's inband data ports where it undergoes up-sampling by a factor equal to the dimensionality of the rotation. Modem transformation results in an orthogonal summation of each up-sample-filtered subsequence into overlapping sub-bands of the transmitter's spectrum. Alternatively, a single spreading function is applied to the information, which is then partitioned according to the desired modulation efficiency into the components of a vector, which is then transformed by a wavelet filter bank. A cooperating receiver filters the input into sub-bands by counter-rotating the signal vector back into data space coordinates. However, the effect of the spreading function must be removed prior to thresholding since the spread signal may have a signal-to-noise ratio of less than 1. “De-spreading” is accomplished by correlating the known spreading functions with the unthresholded data-representation coordinates. Since any vector in the transmitter's data-representation can be re-oriented by an invertible secret transformation (i.e. digitally encrypted by modulo addition of a secret vector) to anywhere in the in-band signal space, a truly secure spread-spectrum can be made. The receiver must decrypt the unthresholded data-representation prior to de-spreading. The receiver counter-rotates, decrypts, de-spreads, and thresholds, in that order. Geometrically, the encrypted vector is always inside a radius R, the radius of the N-dimensional signal space. Modulo-R vector subtraction preserves the large non-integer noise vector added onto the encrypted carrying vector so it may be subsequently correlated (de-spread) and thresholded. Thus, to reiterate, an N-dimensional space is defined by N samples into a D/A converter. The data coordinate system is chosen so that a subset, n of the N coordinates in the data representation, define all the information points that are in-band. For each of the n coordinates, B bits of data can be sent as one of 2^{B }integer levels with an efficiency of 2*B bits/Hz per coordinate (i.e. per sub-band). Since the sub-bands overlap perfectly inside the passband and have typically −70 db stop-bands outside, the overall bandwidth efficiency is nominally 2*B bits/Hz if the filter transition region is negligible compared to the total bandwidth used for data. The remaining N-n data coordinates are not used for data, but can be used for synchronizing the modem. A data vector with integer coordinates is therefore rotated to signal coordinates by a first commuting operation. From a geometric standpoint, this operation can be the application of a polyphase matrix for mapping (or rotation) of a data vector into a signal vector representation. Otherwise, the mapping can be implemented via the use of a Finite Impulse Response (FIR) filter bank or an Infinite Impulse Response (IIP) synthesizer. Alternatively, the mapping can be a wavelet transformation, which can be shown in several cases to be mathematically equivalent to the filtering performed by a synthesizer. The first commuting operator in the transmitter is complimented by a second commuting operator in the receiver. Together, these operators result in an identity matrix, enabling the complete recovery of input data. Practical FIR lattice filter implementations of a Perfect Reconstruction (PR) QMF bank can result in an identity matrix multiplied by a scalar gain and a pure delay. The gain factor arises because the matrices are typically in un-normalized integer format, and the pure, frequency independent delay factor represents the delay through the FIR filters. A Near Perfect Reconstruction (NPR) filter bank is also known to those skilled in the art of QMF design. The FIR lattice QMF is designed by employing the commuting properties of geometric rotations. The FIR transversal filter form is then obtained from the lattice form but in doing so, the PR property is compromised and the filter bank provides NPR. However, by using computer aided design techniques, the NPR filter can usually be optimized to provide an NPR filter having better overall properties for stopband attenuation, for example, than the PR lattice filter used to initiate the iterative design. Therefore, the mappings used to construct the baseband modem modulator and demodulator of the foregoing disclosure and the parent patent application, U.S. Pat. No. 5,367,516, commute in the sense that it may be a perfect or a near-perfect reconstruction with a possible delay and gain. A two-dimensional QMF as in This can be written in matrix form as:
The sample rate change in a QMF combines with the delay column matrix on the right to become a serial/parallel conversion for the analyzer, and the 2×2 matrix on the right, whose elements are shown as row vectors, is the polyphase rotation matrix. The sample rate change also converts the powers of z^{2 }to powers of z, but since only the coefficients of powers of z in the row vectors are relevant to calculations, the z variable is often omitted. A similar definition is used for the QMF synthesizer. The up-sampling at the input to the 2-band synthesizer implies that only the odd coefficients contribute to the odd output sample. Thus, the filtering can proceed at half the rate using filters of half the length, which is what the polyphase notation describes. Polyphase matrices for filters with more than two sub-bands are defined in a similar manner in the cited publications associated with the parent patent, U.S. Pat. No. 5,367,516. The polyphase matrix may be further factored into a form corresponding to an FIR transversal filter that operates on vectors instead of scalars. That is, square matrices C_{j }having scalar elements each multiply the power j power of z. Thus, the modulation operator [M] is described by:
The vector-filter arrangement is of course mathematically equivalent to the other forms of a rotation. However, the transversal form suggests that a modem as described in the foregoing discussion and in the parent patent, U.S. Pat. No. 5,367,516, can be equalized for distortions in the transmission link by using the same methods as are applied to transversal equalizers in the known modem art. Significantly, the equalizer filter and the demodulation filter [D] can be one and the same. Therefore, a further embodiment of the present invention employs adaptive adjustment of the receiver polyphase matrix, also referred to as a de-rotation matrix. If it is determined that some form of frequency dependent distortion is being introduced into the transmission path, the polyphase filter in the receive portion of the present FM modem, or any modem using the methods of as previously discussed and found in the parent patent, U.S. Pat. No. 5,367,516, can be adjusted to compensate for such distortion. Another way of stating this is that some of the de-rotation coefficients in the de-rotation matrix are adjusted to compensate for the frequency-dependent distortion in the transmission path. The adjustment would minimize a computed error function. This is similar to the design procedure for optimizing NPR in the design of the filter bank. In the modem, however, the NPR solution does not accurately reconstruct the information because of amplitude and delay distortions in the modem transmission link. The receiver polyphase filter bank in an FIR implementation is equivalent to FIR filters for each sub-band wherein the filter outputs are decimated and thresholded to recover the data. This is the same form as a fractionally spaced equalizer, which is known to those of ordinary skill in the art of modem equalization. Therefore, the methods in the modem art, such as, but not limited to, the Least Mean Squares (LMS) algorithm, for adaptively adjusting the coefficients of an equalizer in proportion to an error function can be applied to the polyphase matrix coefficients of the demodulator to provide the receiver with an adaptive de-rotation that corrects for link distortions without a separate equalizer filter. Thus, a fractionally sampled FIR filter is provided. The adjustment of the receiver polyphase filter can be done at the initiation of a communication and left at a certain setting, or can be adjusted dynamically by monitoring the RMS spread of noise about the integer or odd-integer threshold values out of the demodulator, as one example. Additive noise in a transmission link between a transmitter and a receiver will add a noise vector to the transmitted signal so that a counter-rotation performed in the receiver will yield a recovered data vector with non-integer coordinates. In a first simplified embodiment of the present invention, a threshold operator takes the nearest integer coordinate values as the most likely symbol. In a more sophisticated receiver embodiment, thresholding is executed according to a Viterbi algorithm since the rotations are convolutional with M-n extra degrees of freedom. This enables “free” error correction without the use of parity bits. In each transmitted frame of a modem according to the foregoing disclosure and the parent patent, U.S. Pat. No. 5,367,516, each D/A sample depends on the scaling input and all other residual inputs to the modulator. To avoid aliasing in the conversion to and from analog by D/A and A/D, the highest frequency sub-band is not used to transmit data in a preferred modem design. Since the number of samples to/from the converter equals the total number of sub-bands M, there is redundancy in the transmitted signal. Other sub-bands may optionally be omitted to shape the output spectrum, for example to avoid transmitting DC, further increasing the redundancy to n out M samples. Sync signals transmitted by the technique of alternating the sign of a fixed amplitude into, for example, the highest frequency sub-bands carry no information and do not increase the redundancy. If the receiver has thresholded a sequence of frames prior to the current frame without bit errors, it may use those results to aid in demodulating the current frame. For example, the unthresholded demodulator output may be midway between two allowed odd-integer threshold values. To decide which soft decision value is most probable, the receiver could use the prior data frames along with the two possible current-frame soft decisions to generate in the receiver a modulated signal to correlate with the actual received signal. Then, the receiver may make a final decision about which integer levels were most probably transmitted for the current frame based upon the better correlation. This procedure, which may be implemented through dynamic programming via the Viterbi Algorithm, is possible because of the redundancy inherent in the convolutional output of the modulator of the foregoing disclosure and the parent patent, U.S. Pat. No. 5,367,516. It differs substantially from the soft forward error correction (FEC) method known in the modem art as Trellis Coded Modulation (TCM). In TCM, as commonly practiced, redundancy is provided by appending one or more suitably calculated parity bits to the data before transmission. The coding gain from the Trellis Coding must make up for the loss in energy per bit due to the transmission of the extra parity. Furthermore, a TCM modem cannot be optimal because of the Shannon capacity sacrificed to send the parity bit. This does not rule out the use of TCM with the current invention. Other methods of forward error correction such as the multi-dimensional coding methods known in the modem art are also applicable to the inherently multi-dimensional signal generated by a coordinate rotation. Thus, a data modem based on rotations that would normally use only odd-integer coordinates for the data vector could instead select the coordinates of successive vectors from a larger set of even and odd integers in a way that decreases the probability of a bit error. Implicit in both the modem and encryptor definitions given above is the assumption that the data will rotate to an N-band analog signal. The N samples of the D/A are within a precise time frame. Placing a strict simultaneous requirement on the bandwidth must be done with care according to known transformation techniques since sine and cosine-based transformations are complicated by the infinite extent of those waveforms. When the time is constrained exactly, the frequency components extend to infinity and vice versa. Wavelet theory provides for transformations based on “mother wavelets” which do not have infinite extent in time. Like Fourier analysis, many basis functions are summed together to represent any arbitrary signal. The mother wavelets are stretched and shifted in time to form a set of residual basis functions and a related set of scaling basis functions. Thus, a single mother wavelet generates a transformation just as a prototype sinewave generates the Fourier transformations. There are enumerable “mother” wavelets, each generating a different transformation. The wavelet transformations are entirely real (there are no imaginary components or carriers). This halves the complexity of modulation and equalization. Since only real components are involved, it is thus possible to frequency modulate an input data stream without having to account for an imaginary component. An exemplary FM modem according to the present disclosure employs an 8-dimensional polyphase filter as the baseband input to an FM modulator. As illustrated in the following table, a non-linear distribution of bits per band is employed.
Thus, in the presently illustrated embodiment, there are a total of 13 bits per input symbol. One design goal is that each band is to have approximately the same power (here, RMS average=9 db). From the following equations, it is apparent that the noise coming out of the FM receiver depends on the square of the sub-band frequency. The power spectral density, or PSD, is given by Couch, “Digital and Analog Communications Systems”, 4th ed., MacMillan, equation 7-125:
According to this frequency versus noise relationship, fewer and fewer bits may be transmitted within each sub-band as the sub-bands move away from DC since the noise increases as the square of the frequency. Therefore, it is possible to send a larger number of bits at the lowest frequencies since there is significantly less noise coming out of the discriminator within the receiver then at the higher sub-bands. In the above example, sub-band 0 has 4 bits, while sub-band 6 has 1 bit; the number of levels in each band are chosen to match the parabolic noise density function, PSD, of an FM discriminator. It is possible in a first embodiment of the present invention to provide to an FM modulator sub-bands having decreasing average power levels with increasing sub-band frequency. For example, the higher sub-bands can carry fewer bits but all sub-bands use the same level spacing per bit. In a second embodiment of the present invention, the levels representing each bit or bits in the sub-bands are spaced apart or pre-emphasized to provide equivalent average power across all sub-bands. In other words, voltage levels of +/−1, 3, 5, 7, 9, 11, 13 and 15 are employed in sub-band 0 in order to represent the data conveyed by the 4 bits assigned to this sub-band. Voltage levels of +/−9 are used to represent the 0 or 1 state of the lone bit assigned to sub-band 6. The average power for these two sub-bands is thus approximately the same. This technique of spacing the levels apart at higher frequency sub-bands is referred to as a pre-emphasis technique. With a signal pre-emphasized in this manner in its data coordinate representation, it can then be converted to its signal coordinate representation by a polyphase coordinate rotation filter. This sampled analog signal can then be used as an input to an analog FM transmitter. A receiver detects these levels and converts them back to bits. Thus, no explicit pre-emphasis or de-emphasis filters are needed. In the above example, the levels for each sub-band are given as integers (i.e. +/−1, 3, 5, . . . ). In other words, each sub-band has 2^{B }levels. In a further embodiment of the present application, the number of levels in a given sub-band is not a power of two, but if a total of K bits are transmitted in p sub-bands, then the total number of levels in p sub-bands is 2^{K}. A binary mapping algorithm is then employed for determining which of the two to the n levels is represented. Adjusting the bits per sub-band and total number of bits per symbol enables the optimization of bandwidth for carrier to noise ratio. Computer aided design iterations provide these optimized values. In order to compensate for the pre-emphasis of the higher frequency sub-bands, it is necessary to “de-emphasize” the received signal at the higher frequencies in the receiver. Less data is sent in higher frequency sub-bands, though the data is conveyed at higher levels to provide uniform power out of the transmitter across the sub-bands. Attenuating the high frequency sub-bands results in de-emphasis gain Gd. This de-emphasis results in an overall gain which, for M dimensions, may be approximated by the following formula:
Designating H(i) as the pre-emphasis amplification in sub-band i:
where the denominator in the preceding expression for Gs is proportional to the integral from subband i to subband i+1 of the PSD of the noise out of the discriminator. Sub-band i=0 carries k bits of data and no pre-emphasis so H(0)=1. Sub-band 1 carries k−1 bits with H(1)=2, etc. A typical assignment for M=8 results in nearly equal power in each sub-band:
for a total of B=26 bits per symbol. The various bit level assignments are as illustrated above in the previous example of bits per sub-band. The total gain G is derived from the de-emphasis gain Gd, as given above, the normal FM transmitter gain Gm (also known as the modulation gain), which is due to the FM index, as given below, and the noise-reduction gain Gr. The total gain G is given by:
The FM index is the ratio of the peak carrier frequency deviation to the peak frequency of the baseband signal. As shown in the foregoing formulas, as the peak-to-average ratio (PAR) increases, the FM gain, or modulation gain, decreases. To address this problem, a further embodiment of the present invention employs non-linear amplification of the baseband signal prior to insertion into the FM transmitter and complimentary de-amplification at the receiver, collectively referred to as companding. In a preferred embodiment, a logarithmic Mu-law function performs the transmitter amplification digitally by table look-up. The Mu-law is applied to the signal coordinate representation. In an embodiment preferred for large carrier to noise ratios (CNR), the data bits are framed and expressed in data coordinates, then pre-emphasized, then rotated, then Mu-law amplified by look-up before the application to the FM modulator as a sampled sequence. As previously noted, in one embodiment, the de-compression attenuation includes an equalizer. In another embodiment preferred for small CNR, the data is framed and expressed in data coordinates, rotated to signal coordinates, digitally Mu-law amplified, rotated back to data coordinates, pre-emphasized, rotated to signal coordinates and then applied to the FM modulator. This latter configuration, which permits de-emphasis immediately after the discriminator, may be preferred when operating near FM threshold or when operating in the presence of high-amplitude non-Guassian interference. The Mu-law function preferred for the transmitter non-linear amplification is:
A second benefit of logarithmic amplification is a result of the necessary logarithmic de-amplification (i.e. de-compression) at the receiver, which amounts to attenuation. The input signal has been logarithmically amplified and then de-amplified at either ends of a transmission link. However, noise has been introduced within this transmission link. Therefore, this noise is inverse logarithmically attenuated, resulting in the noise reduction gain Gr, as shown in the receiver gain equation above. The bit error rate of a modem depends upon the energy per bit, or in other words, the energy separating each data representation level. This rate also depends upon the noise energy density. The signal to noise ratio of an FM modem is:
The signal to noise ratio of a modem is Eb/N0 times the bandwidth efficiency of the modem (given in bits per second per hertz), or:
The foregoing application of FM transmission techniques to data transmission derives benefit from the fact that amplitude distortion is irrelevant for FM since information is retrieved by the receiver from the frequency by, for example, tracking zero-crossings. In It is commonly recognized that frequency modulation can be achieved either by: 1) directly adjusting the frequency of a carrier; or 2) adjusting the phase of a carrier with a signal which is the integral of the information to be transmitted. The use of phase adjustment to achieve frequency modulation is preferred in one embodiment of the present invention since chips are available which enable phase modulation in an entirely digital fashion, thus maximizing the reproducibility of the input signal. Phase adjustment FM is illustrated in In FM double side-band provides significant advantages over prior art modems. It provides immunity to noise associated with amplitude modulated signals. Distortion due to non-linear amplification is avoided. The RF amplifier employed need not be as precise (and expensive) since amplitude modulation need not be accurately reproduced. It also provides a certain degree of circuit simplicity, since carrier recovery is not necessary at the receiver. However, FM double side-band provides half the band-width efficiency as compared to FM single side-band. Therefore, another embodiment of the present invention employs FM single side-band. The flow diagram of A Hilbert Transform processes the input signal, effectively phase shifting the input by 90 degrees (step 254). The output of the Hilbert filtration is likewise integrated (step 256). The Hilbert filter integration is then entered into an exponential lookup table (step 258). A sinusoidal carrier signal is digitally simulated as in step 260 via the use of a lookup table. This carrier is then phase modulated using the integrated polyphase filter output (step 262). As with the FM double side-band scenario, the zero crossings of the phase modulated carrier are computed (step 264), and a digital pulse train corresponding to the zero crossings is produced (step 266). Finally, in the single side-band case, a switching amplifier whose input current is proportional to the output from the exponential look-up table output amplifies the digital zero crossing pulse train (step 268). In either the FM double or single side-band cases, the first steps in the transmission procedure, polyphase filtering the input data, is achieved as follows. With reference to A stack is employed in step 272 is order to retain this and L previous vectors. As indicated in the figure, a first embodiment uses L=9 vectors. These L vectors are multiplied by an M×M dimensional scalar matrix to effectuate the initial polyphase filtering (step 274). As illustrated, the scalar matrix is 8×8. Subsequently, in step 276, the matrix multiplication occurs on the (L−1) previously stored vectors in the stack using different 8×8 matrices for each iteration, and the result is accumulated and output (step 278). As noted in step 280, the previous steps shown in In all of the foregoing illustrations, it is understood that alternative procedures for FM modulating input data according to the present polyphase technique are feasible. However, the illustrated procedure maximizes the proportion of data manipulations which are digital. With reference to The DC filtration of step 290 is useful in applications of the present invention which are subject to substantial doppler shift. For modems used for low earth orbiting satellites, where the doppler shift follows known orbital parameters, filtering is used in one example in order to eliminate the effects of such DC or low frequency distortion. Another approach to addressing such low frequency distortion is to avoid use of the lowest frequency sub-band. The decimation filtering of the same step effects a smoothing of the zero crossing count, the latter necessarily being at a high rate for adequate signal resolution, but which may introduce unwanted noise. In step 292, polyphase filtration involves the application of the commuting operator employed in the first step of the transmission procedure. Thresholding is used to differentiate between received levels. The transmitted signal was sent at one of multiple, easily distinguished levels. However, noise introduced in the transmission link shifts the received signals to points intermediate the expected levels. It is therefore necessary in step 294 to assign the decimated receive signal to one of multiple signal levels. Though carrier recovery is not required using FM data transmission, baseband synchronization requires some form of bit sync recovery (step 294). As previously noted, such sync information can be transmitted in sub-bands not employed for data transmission. If doppler shift is expected to be a problem, such as in low earth orbit satellite communications, the lowest frequency sub-band may not be used for data, and may be available for sync transmission. Similarly, the highest frequency sub-band cannot typically be used. A filter perfect enough to avoid aliasing in the equivalent A to D and D to A conversion operations does not exist. By transmitting a known bit pattern at the 3 db point between the second to highest sub-band and the highest sub-band, the sync information is within the bandwidth of the transmitted signal, but is not taking up usable data bandwidth. Such sync information can be inserted by sampling a DC signal at a sync bit rate. High and low pass filtering result in a sinewave output at the sample rate, each square wave being at the high or low end of the respective sub-band. Thus, this sync signal, akin to a pilot tone, is orthogonal to all of the data signals. The specific steps involved in the application of the operator of step 292 is illustrated in The aforementioned steps illustrate that this is, for all practical purposes, an all digital FM modem. The only non-digital parts are a passive tank circuit employed in steps 244 and 268, the receiver front end in steps 284 and 286, and a D to A converter necessary to modulate the input current to the Class C switching amplifier using the output of the exponential lookup table. The transmitter RF power amplifier is in fact a digital switch. The filters could also be implemented so that all multiplications become binary shifts and adds, thus reducing the cost and power dissipation of an ASIC. The previous description and equations generally assume that only one of M subbands is unused to prevent aliasing in analog converters. In alternative embodiments, a speed advantage may flow from allowing more than one subband to be unused. As previously recognized, the instant technique and general structure can be adapted in a further embodiment for the transmission of polyphase filtered data using AM modulation. Adaptive rotation of the receive rotation matrix, as previously described, can be applied to an AM modulated modem which employs the LMS algorithm. For the LMS method, the rows of the vector filter tap matrices are corrected iteratively by subtracting a vector proportional to the error times the input vector stored in the filter delay line for that tap. The error can be either the difference between the unquantized and quantized output of the modem receiver (i.e. the error margin), or the error can be the difference between the unquantized output and a known training data sequence. These general techniques are like those used in modem equalizers except they are applied to the rotation matrix without a separate equalizer. One embodiment of the foregoing modulation and demodulation technique is modelled using the software attached hereto as an Appendix. Having described preferred embodiments of the invention, it will now become apparent to one of ordinary skill in the art that other embodiments incorporating the concepts may be used. It is felt, therefore, that these embodiments should not be limited to disclosed embodiments but rather should be limited only by the spirit and scope of the appended claims. Referenced by
Classifications
Legal Events
Rotate |