US 20020135502 A1 Abstract A method and apparatus for convolution encoding and Viterbi decoding utilizes a flexible, digital signal processing architecture that comprises a core processor and a plurality of re-configurable processing elements arranged in a two-dimensional array. The core processor is operable to configure the re-configurable processing elements to perform data encoding and data decoding functions. A received data input is encoded by configuring one of the re-configurable processing elements to emulate a convolution encoding algorithm and applying the received data input to the convolution encoding algorithm. A received encoded data input is decoded by configuring the plurality of re-configurable processing elements to emulate a Viterbi decoding algorithm wherein the plurality of re-configurable processing elements is configured to accommodate every data state of the convolution encoding algorithm. The core processor initializes the re-configurable processing elements by assigning register values to registers that define parameters such as constraint length and code rate for the convolution encoding algorithm.
Claims(20) 1. In a digital signal processing architecture comprising a core processor and a plurality of re-configurable processing elements arranged in a two-dimensional array, a method for processing digital data comprises the steps of:
encoding a received data input by configuring one of said re-configurable processing elements to emulate a convolution encoding algorithm and applying said received data input to said convolution encoding algorithm to produce an encoded data output; and decoding a received encoded data input by configuring said plurality of re-configurable processing elements to emulate a Viterbi decoding algorithm, wherein said plurality of re-configurable processing elements is configured to accommodate every data state of said convolution encoding algorithm, and applying said received encoded data input to said Viterbi decoding algorithm to produce a decoded data output. 2. The method of 3. The method of 4. The method of 5. The method of 6. The method of 7. The method of 8. The method of 9. The method of 10. The method of 11. A digital signal processing architecture comprising:
a plurality of re-configurable processing elements arranged in a two-dimensional array, said plurality of re-configurable processing elements are programmable in response to predetermined context words; and a processor operatively coupled to said plurality of re-configurable processing elements to control loading of said predetermined context words thereto, said processor being selectively operable to configure one of said re-configurable processing elements to emulate a convolution encoding algorithm, and to configure said plurality of re-configurable processing elements to emulate a Viterbi decoding algorithm wherein said plurality of re-configurable processing elements accommodates every data state of said convolution encoding algorithm; wherein, a received data input applied to said convolution encoding algorithm produces an encoded data output, and a received encoded data input applied to said Viterbi decoding algorithm produces a decoded data output. 12. The digital signal processing architecture of 13. The digital signal processing architecture of 14. The digital signal processing architecture of 15. The digital signal processing architecture of 16. The digital signal processing architecture of 17. The digital signal processing architecture of 18. The digital signal processing architecture of 19. The digital signal processing architecture of 20. The digital signal processing architecture of Description [0001] 1. Field of the Invention [0002] The present invention relates to digital signal processing, and more particularly to the mapping of a convolution encoder and a Viterbi decoder onto a dynamically re-configurable two-dimensional single instruction multiple data (SIMD) processor array architecture. [0003] 2. Description of Related Art [0004] The field of digital signal processing (DSP) has grown dramatically in recent years and has quickly become a key component in many consumer, communications, medical, and industrial products. DSP technology involves the analyzing and processing of digital data in the form of sequences of ones and zeros. In the field of communications, analog signals are converted to such digital sequences for processing and transmission. During transmission, however, these digital sequences may be easily distorted by noise. In order to address this problem, digital data is often encoded before transmission. One form of encoding, known as convolution encoding, is widely used in digital communication and signal processing to protect transmitted data against noise, and its efficiency is well known in terms of error correction quality. In general, convolution encoding is a coding scheme that associates at least one encoded data element with each source data element to be encoded, this encoded data element being obtained by the modulo-two summation of this source data element with at least one of the previous source data elements. Thus, each encoded symbol is a linear combination of the source data element to be encoded and the previous source data elements. [0005] In FIG. 1A, a schematic diagram of a standard convolution encoder with a code rate of one half is shown. For this type of encoder, two encoding outputs, a(t) and b(t), are transmitted for every input u(t). The encoder is shown to be comprised of two delay elements, [0006] The encoding process of the described encoder may also be characterized by the finite state machine illustrated in FIG. 1B. In this diagram, each circle is labeled with a binary representation of one of the four binary states of the encoder. In particular, this diagram provides binary representations for state zero [0007] For example, beginning at state zero [0008] In order to depict how the described encoder evolves over time, a trellis diagram is shown in FIG. 1C. As illustrated, this diagram is comprised of several nodes (denoted by dots) and transition paths (denoted by solid lines). Each column of nodes represents all states at a particular instant. In this particular example, five instants are described (corresponding to t=1 through t=5). Therefore, this trellis diagram can be regarded as illustrating the sequence of all possible state transition paths over five instants (where it is assumed that the initial state is state zero [0009] In general, noise will occur during transmission. For example, if the observed output sequence is { [0010] In the presence of noise, the most commonly used approach to decode convolution codes is via the Viterbi algorithm. In particular, the Viterbi algorithm gives a binary estimation of each input u(t) coded at transmission. This estimation is determined by finding the most likely transition path of a given trellis with respect to the noisy output data (X(t), Y(t)) received by a decoder respectively corresponding to the originally encoded output data (a(t), b(t)). Each node of the trellis used during decoding contains an information element on the survivor path of the two possible paths ending at that particular node. The basic principle of the Viterbi algorithm consists in considering, at each node, only the most probable path as to enable easy trace-back operations on the trellis and hence to determine an a posteriori estimation of the value received several reception instants earlier. [0011] The Viterbi algorithm involves the execution of a particular set of operations. First, a computation is made of the distances, also called branch metrics, between the received noisy output data (X(t), Y(t)) and the symbols (a(t), b(t)) corresponding to the required noiseless outputs of a particular state transition path. In particular these branch metric units are defined as: Branch( [0012] where (a Branch ( Branch ( [0013] where Branch ( [0014] where P(j, t) represents the cumulative branch metric of state j at instant t, P(i, t−1) represents the cumulative branch metric of a state i preceding state j at instant (t−1), and Branch [0015] where {i*} represents the set of states having transitions into state j. It should be noted that the above formula is only needed when there are two possible state transition paths into a particular node (otherwise, the most likely path into state j M(j, t) is simply P(j, t)). In the current example, it should thus be clear that this calculation is not needed until the fourth instant (t=4). It should also be noted that, in the current example, it is assumed that all cumulative branch metrics are initially zero. Therefore, P( [0016] In the next instant (t=2), four branch metric calculations are needed. Namely, the following branches are needed: Branch ( Branch ( Branch ( Branch ( [0017] The cumulative branch metrics corresponding to the two possible paths for each state are then compared in order to determine the paths most likely taken at this particular instant. The selected paths and the cumulative branch metrics of each state are then both stored in memory until the next instant. [0018] After a pre-determined number of instants, a trace-back operation is made in order to determine the optimal cumulative path taken. In particular, the path with the largest cumulative path metric is chosen as the optimal path (although some implementations use the smallest cumulative path metric). This optimal path is then used to decode the original coded bit stream of information according the procedure described earlier for noiseless conditions. [0019] The Viterbi algorithm has been implemented in the prior art using either hardware or software systems. Software implementations of the Viterbi algorithm adapted to run on general purpose digital signal processors have the advantage of better flexibility than hardware implementations, since the software can be readily reprogrammed. Conversely, hardware implementations of the Viterbi algorithm using application specific integrated circuits (ASICs) can achieve higher performance than the software implementations in terms of lower power consumption, higher decoding rates, etc., but cannot be easily modified. [0020] It would therefore be advantageous to develop a method and apparatus for convolution encoding and Viterbi decoding that addresses these limitations of known hardware and software implementations. More specifically, it would be advantageous to develop a method and apparatus for convolution encoding and Viterbi decoding that has the flexibility of the software implementations, with the superior performance of the hardware implementations. [0021] A method and apparatus for convolution encoding and Viterbi decoding utilizes a flexible, digital signal processing architecture that comprises a core processor and a plurality of re-configurable processing elements arranged in a two-dimensional array. The present invention therefore enables the convolution encoding and Viterbi decoding functions to be mapped onto this flexible architecture, thereby overcoming the disadvantages of conventional hardware and software solutions. [0022] In an embodiment of the invention, the core processor is operable to configure the re-configurable processing elements to perform data encoding and data decoding functions. A received data input is encoded by configuring one of the re-configurable processing elements to emulate a convolution encoding algorithm and applying the received data input to the convolution encoding algorithm. A received encoded data input is decoded by configuring the plurality of re-configurable processing elements to emulate a Viterbi decoding algorithm wherein the plurality of re-configurable processing elements is configured to accommodate every data state of the convolution encoding algorithm. The core processor initializes the re-configurable processing elements by assigning register values to registers that define parameters such as constraint length and code rate for the convolution encoding algorithm. [0023] More particularly, the encoding function further comprises generating a multiple output sequence corresponding to the received data input. Essentially, the encoding function comprises performing a modulo-two addition of selected taps of a serially timedelayed sequence of the received data input. The decoding function further comprises mapping a trellis diagram onto the plurality of re-configurable processing elements. The re-configurable processing elements calculate cumulative branch metric units for each node of the trellis diagram, and the core processor selects a most probable state transition path of the trellis diagram based on the branch metric units. [0024] A more complete understanding of the method and apparatus for convolution encoding and Viterbi decoding will be afforded to those skilled in the art, as well as a realization of additional advantages and objects thereof, by a consideration of the following detailed description of the preferred embodiment. Reference will be made to the appended sheets of drawings which will first be described briefly. [0025]FIG. 1A is a schematic diagram of a convolution encoder having a code rate of one half; [0026]FIG. 1B is a schematic diagram of a finite state machine of an encoder having a code rate of one half; [0027]FIG. 1C is a trellis diagram illustrating the possible state transitions of encoded data having a code rate of one half; [0028]FIG. 2 is a block diagram of a preferred embodiment of the invention; [0029]FIG. 3A is a schematic diagram illustrating the internal quadrants of the RC array; [0030]FIG. 3B is a schematic diagram illustrating the internal express lanes of the RC array; [0031]FIG. 3C is a schematic diagram illustrating the internal data-bus connections of the RC array; [0032]FIG. 4A is a schematic diagram of a convolution encoder having a code rate of one third and constraint length of nine; [0033]FIG. 4B is a trellis diagram illustrating the possible state transitions of encoded data having a code rate of one third and constraint length of nine; [0034]FIG. 5 is a diagram illustrating the various registers allocated for encoding in an RC; [0035]FIG. 6 is a flow chart illustrating the steps for encoding one bit of information according to a preferred embodiment of the invention; [0036]FIG. 7 is a flow chart illustrating the steps for decoding a bit stream of information according to a preferred embodiment of the invention; [0037]FIG. 8 is diagram illustrating the state transition mapping of a Viterbi decoder for encoded data having a code rate of one third and a constraint length of nine; [0038]FIG. 9 is a diagram illustrating the branch metric mapping of a Viterbi decoder for encoded data having a code rate of one third and a constraint length of nine; and [0039]FIG. 10 is a schematic diagram demonstrating the data collapse procedure for writing path information into the frame buffer. DE [0040] The present invention is directed towards a method and apparatus for convolution encoding and Viterbi decoding. In particular, this invention provides a unique re-configurable architecture that addresses the performance limitations currently known in the art by simultaneously achieving the flexibility of software pertaining to general-purpose processors and sustaining the high performance pertaining to hardware implementations of application-specific circuits. In the detailed description that follows, it should be appreciated that like element numerals are used to describe like elements illustrated in one or more of the figures. [0041] An embodiment of the invention shown in FIG. 2 comprises an architecture including a dynamically re-configurable two-dimensional SIMD processor array [0042] In a preferred embodiment of this invention, the core processor [0043] In FIG. 3A, a diagram illustrating the internal connections of the RC array [0044] Returning to the architecture illustrated in FIG. 2, the function of each component is now described. The processing element of this invention is called the re-configurable cell (RC) [0045] The programmability of this architecture is derived from context words that are broadcast to the rows or columns of the RC array [0046] A method in accordance with an embodiment of this invention is described for the case of a standard convolutional code, with a constraint length of nine and a code rate of one third, obtained by means of an exemplary coder shown in FIG. 4A. It should be understood that the decoding method and apparatus presented by this invention may be applied to all convolutional codes having code rates of η=1/K (where K is an integer >1) and varying constraint lengths, by a simple generalization of the described method. As illustrated, convolution encoding involves the modulo-two addition of selected taps of a serially time-delayed data sequence. As illustrated in FIG. 4A, an input u(t) is passed through a series of eight delay elements [0047] The dynamics of this coder are described by the diagram of the trellis shown in FIG. 4B and are well known in the art. For this particular example, it is shown that for each of the two hundred fifty six possible current states, there are two potential state transition paths that can be taken into the next state. For example, if a zero input u(t) is passed through the coder when the current state is zero (S [0048] In a preferred embodiment of the present invention, only one RC [0049] In FIG. 6, a flow chart describing the encoding procedure for one bit of data is provided. Encoding begins at step [0050] In FIG. 7, a flow chart illustrating the steps for decoding a bit-stream of encoded data is shown. For simplicity, the mapping of the Viterbi decoder onto the aforementioned RC array [0051] Once this first packet of data is loaded into the RC array Branch( [0052] where −a [0053] Next, the selected path is recorded and written back to the frame buffer [0054] As illustrated, this process begins by taking the first two bits of each RC [0055] Returning to the flow chart illustrated in FIG. 7, a re-ordering of the state metrics is then made at step [0056] After updating these branch metric values at step [0057] Having thus described a preferred embodiment of the method and apparatus for convolution encoding and Viterbi decoding, it should be apparent to those skilled in the art that certain advantages of the within system have been achieved. It should also be appreciated that various modifications, adaptations, and alternative embodiments thereof may be made within the scope and spirit of the present invention. The invention is further defined by the following claims. Referenced by
Classifications
Legal Events
Rotate |