Chip design for high-performance DSP

Advances in silicon technology have been a key development in the realisation of many telecommunication and signal processing systems. In many cases, the development of application-specific digital signal processing (DSP) chips is the most cost-effective solution and provides the highest performance. Advances made in computer-aided design (CAD) tools and design methodologies now allow designers to develop complex chips within months or even weeks. This paper gives an insight into the challenges and design methodologies of implementing advanced highperformance chips for DSP. In particular, the paper reviews some of the techniques used to develop circuit architectures from high-level descriptions and the tools which are then used to realise silicon layout.

CLOSE—A Data-Driven Approach to Speech Separation

This paper studies single-channel speech separation, assuming unknown, arbitrary temporal dynamics for the speech signals to be separated. A data-driven approach is described, which matches each mixed speech segment against a composite training segment to separate the underlying clean speech segments. To advance the separation accuracy, the new approach seeks and separates the longest mixed speech segments with matching composite training segments. Lengthening the mixed speech segments to match reduces the uncertainty of the constituent training segments, and hence the error of separation. For convenience, we call the new approach Composition of Longest Segments, or CLOSE. The CLOSE method includes a data-driven approach to model long-range temporal dynamics of speech signals, and a statistical approach to identify the longest mixed speech segments with matching composite training segments. Experiments are conducted on the Wall Street Journal database, for separating mixtures of two simultaneous large-vocabulary speech utterances spoken by two different speakers. The results are evaluated using various objective and subjective measures, including the challenge of large-vocabulary continuous speech recognition. It is shown that the new separation approach leads to significant improvement in all these measures.

Valved Dataflow For FPGA Memory Hierarchy Synthesis

For modern FPGA, implementation of memory intensive processing applications such as high end image and video processing systems necessitates manual design of complex multilevel memory hierarchies incorporating off-chip DDR and onchip BRAM and LUT RAM. In fact, automated synthesis of multi-level memory hierarchies is an open problem facing high level synthesis technologies for FPGA devices. In this paper we describe the first automated solution to this problem.
By exploiting a novel dataflow application modelling dialect, known as Valved Dataflow, we show for the first time how, not only can such architectures be automatically derived, but also that the resulting implementations support real-time processing for current image processing application standards such as H.264. We demonstrate the viability of this approach by reporting the performance and cost of hierarchies automatically generated for Motion Estimation, Matrix Multiplication and Sobel Edge Detection applications on Virtex-5 FPGA.

From bit level systolic arrays to HDTV processor chips

The initial part of this paper reviews the early challenges (c 1980) in achieving real-time silicon implementations of DSP computations. In particular, it discusses research on application specific architectures, including bit level systolic circuits that led to important advances in achieving the DSP performance levels then required. These were many orders of magnitude greater than those achievable using programmable (including early DSP) processors, and were demonstrated through the design of commercial digital correlator and digital filter chips. As is discussed, an important challenge was the application of these concepts to recursive computations as occur, for example, in Infinite Impulse Response (IIR) filters. An important breakthrough was to show how fine grained pipelining can be used if arithmetic is performed most significant bit (msb) first. This can be achieved using redundant number systems, including carry-save arithmetic. This research and its practical benefits were again demonstrated through a number of novel IIR filter chip designs which at the time, exhibited performance much greater than previous solutions. The architectural insights gained coupled with the regular nature of many DSP and video processing computations also provided the foundation for new methods for the rapid design and synthesis of complex DSP System-on-Chip (SoC), Intellectual Property (IP) cores. This included the creation of a wide portfolio of commercial SoC video compression cores (MPEG2, MPEG4, H.264) for very high performance applications ranging from cell phones to High Definition TV (HDTV). The work provided the foundation for systematic methodologies, tools and design flows including high-level design optimizations based on "algorithmic engineering" and also led to the creation of the Abhainn tool environment for the design of complex heterogeneous DSP platforms comprising processors and multiple FPGAs. The paper concludes with a discussion of the problems faced by designers in developing complex DSP systems using current SoC technology. © 2007 Springer Science+Business Media, LLC.

VLSI architectures for vector quantization

The real time implementation of an efficient signal compression technique, Vector Quantization (VQ), is of great importance to many digital signal coding applications. In this paper, we describe a new family of bit level systolic VLSI architectures which offer an attractive solution to this problem. These architectures are based on a bit serial, word parallel approach and high performance and efficiency can be achieved for VQ applications of a wide range of bandwidths. Compared with their bit parallel counterparts, these bit serial circuits provide better alternatives for VQ implementations in terms of performance and cost. © 1995 Kluwer Academic Publishers.

High performance VLSI architecture for Wave Digital Filtering

The application of fine grain pipelining techniques in the design of high performance Wave Digital Filters (WDFs) is described. It is shown that significant increases in the sampling rate of bit parallel circuits can be achieved using most significant bit (msb) first arithmetic. A novel VLSI architecture for implementing two-port adaptor circuits is described which embodies these ideas. The circuit in question is highly regular, uses msb first arithmetic and is implemented using simple carry-save adders. © 1992 Kluwer Academic Publishers.

Bit-Level systolic architectures for high performance IIR filtering

Several novel systolic architectures for implementing densely pipelined bit parallel IIR filter sections are presented. The fundamental problem of latency in the feedback loop is overcome by employing redundant arithmetic in combination with bit-level feedback, allowing a basic first-order section to achieve a wordlength-independent latency of only two clock cycles. This is extended to produce a building block from which higher order sections can be constructed. The architecture is then refined by combining the use of both conventional and redundant arithmetic, resulting in two new structures offering substantial hardware savings over the original design. In contrast to alternative techniques, bit-level pipelinability is achieved with no net cost in hardware. © 1989 Kluwer Academic Publishers.

A bit-level systolic architecture for implementing a VQ tree search

A bit-level systolic array system for performing a binary tree Vector Quantization codebook search is described. This consists of a linear chain of regular VLSI building blocks and exhibits data rates suitable for a wide range of real-time applications. A technique is described which reduces the computation required at each node in the binary tree to that of a single inner product operation. This method applies to all the common distortion measures (including the Euclidean distance, the Weighted Euclidean distance and the Itakura-Saito distortion measure) and significantly reduces the hardware required to implement the tree search system. © 1990 Kluwer Academic Publishers.

Fast VLSI algorithms for division and square root

Real time digital signal processing demands high performance implementations of division and square root. This can only be achieved by the design of fast and efficient arithmetic algorithms which address practical VLSI architectural design issues. In this paper, new algorithms for division and square root are described. The new schemes are based on pre-scaling the operands and modifying the classical SRT method such that the result digits and the remainders are computed concurrently and the computations in adjacent rows are overlapped. Consequently, their performance exceeds that of the SRT methods. The hardware cost for higher radices is considerably more than that of the SRT methods but for many applications, this is not prohibitive. A system of equations is presented which enables both an analysis of the method for any radix and the parameters of implementations to be easily determined. This is illustrated for the case of radix 2 and radix 4. In addition, a highly regular array architecture combining the division and square root method is described. © 1994 Kluwer Academic Publishers.

QRD and SVD processor design based on an approximate rotations algorithm

A silicon implementation of the Approximate Rotations algorithm capable of carrying the computational load of algorithms such as QRD and SVD, within the real-time realisation of applications such as Adaptive Beamforming, is described. A modification to the original Approximate Rotations algorithm to simplify the method of optimal angle selection is proposed. Analysis shows that fewer iterations of the Approximate Rotations algorithm are required compared with the conventional CORDIC algorithm to achieve similar degrees of accuracy. The silicon design studies undertaken provide direct practical evidence of superior performance with the Approximate Rotations algorithm, requiring approximately 40% of the total computation time of the conventional CORDIC algorithm, for a similar silicon area cost. © 2004 IEEE.

Rijndael FPGA implementations utilising look-up tables

This paper presents single-chip FPGA Rijndael algorithm implementations of the Advanced Encryption Standard (AES) algorithm, Rijndael. In particular, the designs utilise look-up tables to implement the entire Rijndael Round function. A comparison is provided between these designs and similar existing implementations. Hardware implementations of encryption algorithms prove much faster than equivalent software implementations and since there is a need to perform encryption on data in real time, speed is very important. In particular, Field Programmable Gate Arrays (FPGAs) are well suited to encryption implementations due to their flexibility and an architecture, which can be exploited to accommodate typical encryption transformations. In this paper, a Look-Up Table (LUT) methodology is introduced where complex and slow operations are replaced by simple LUTs. A LUT-based fully pipelined Rijndael implementation is described which has a pre-placement performance of 12 Gbits/sec, which is a factor 1.2 times faster than an alternative design in which look-up tables are utilised to implement only one of the Round function transformations, and 6 times faster than other previous single-chip implementations. Iterative Rijndael implementations based on the Look-Up-Table design approach are also discussed and prove faster than typical iterative implementations.

Synthesizable FFT cores

Methods are presented for developing synthesizable FFT cores. These are based on a modular approach in which parameterizable blocks are cascaded to implement the computations required across a range of typical FFT signal flow graphs. The underlying architectural approach combines the use of a digital serial data organization with generic commutator blocks to produce systems that offer 100% processor utilization with storage requirements less than previous designs. The approach has been used to create generators for the automated synthesis of FFT cores that are portable across a broad range of silicon technologies. Resulting chip designs are competitive with manual methods but with significant reductions in design times.