This survey of the development of the modern digital signal processor has concentrated on the mainstream DSP developments which have given us incredibly powerful and low-cost arithmetic processors. The view presented here is that the modern DSP is not historically derived from the conventional general purpose von Neumann processor, but from a separate 'species' of digital signal processors. Any similarities to the general purpose processor are of quite recent origin, driven by market forces and only skin deep. The DSP should in no way be seen as an enhanced performance general purpose microprocessor. The concentration on filtering, FFT and similar applications has meant that electronic engineers, rather than computer scientists have con trolled DSP development so that software developments have often lagged behind the von Neumann machines. Nevertheless, mainstream DSP development seems to be in the process of implementing the lessons learned from other computing activities.

In this chapter the z-transform representation of discrete signals and systems has been developed. The basic concepts have been reviewed, and the use of the transform in evaluating the frequency response has been shown. It is seen that the z-transform method of solving difference equations is analogous to the Laplace transform method; i.e. both techniques first transform the equations to polynomial representations in the corresponding complex variable and then invert them to obtain a closed-form solution.

This chapter considers the radix-2 FFT algorithm where the total number of data samples is an integer factor of two and discusses its application to spectral analysis of signals.

Non-recursive digital filters have a finite impulse response (FIR) sequence, and they are inherently stable. Furthermore, a digital filter with a symmetrical impulse response has a linear phase characteristic, and therefore in this case there is no phase distortion imposed by the filter. In this chapter FIR filter design will be illustrated by considering the moving-averager filter, the frequency sampling method of design, and frequency-domain filter design using window functions.

It has been shown that quantisation and rounding errors sometimes have considerable effects on the performance of digital filters, and therefore it is important that these are taken into account when considering their practical implementation. In this respect, to achieve adequate precision for the signal and number representation, a good estimate of the required processor word length must be made, and, if possible, a safety factor should be used so that three or four bits are added to the estimated word length.

This chapter presents an elementary introduction to Kalman filtering, starting from the simplest of all estimation problems, namely that of estimating a time independent scalar quantity from a number of noisy measurements. From this it move on to consider the case when the quantity to be estimated is a function of time, and then generalise the results to the estimation of a time dependent vector. Finally it indicates how the resulting Kalman filter equations can be applied to an elementary but nevertheless real problem in navigation.

The use of digital signal processors as controllers offers a number of potential advantages. In particular DSPs incorporate a high degree of parallelism enabling the use of digital control in some very demanding applications. Areas such as robot control, turbogenerators and flight control systems immediately suggest themselves. In this context it is relevant to note that traditional 8-bit microprocessors can execute a 2nd order controller algorithm (including “housekeeping”) in about 30 ms while more modern devices are capable of the same computation in 1ms. DSP devices can, by contrast, achieve times as short as 400ns.

As DSP devices become more complex, they are finding their way into much larger, system level applications requiring inter-processor communications and realtime multitasking operating systems; as lower performance devices become cheaper their use in medium to low cost, high performance industrial control applications is increasing. In most DSP applications, sophisticated tool sets for system design and development are almost essential to allow DSP users to get high performance products to the market place in the shortest possible time. Simulation tools are becoming an increasingly important part of the tool set, improving productivity at both the system design stage and in the development of software.

The most widely used type of digital signal processor is the stored program monolithic processor, used for general purpose signal processing and other applications requiring fast arithmetic. These three DSPs have some features in common, such as fast, synchronous serial ports, special addressing modes and on-chip memory.

In this chapter attention has been focused mainly on research undertaken by the author and his colleagues. Considerable effort has also been devoted in many other laboratories world-wide on research on VLSI array processor architectures which are pipelined at the bit level and suitable for high performance DSP chip design. Quite a number of these architectures have been used as the basis of chip designs. Further information on these designs are available from a number of sources.

This chapter presents a case study of computer-aided instrumentation for on-line measurement and control in the area of critical care medicine. The system has been initially developed to test the hypothesis that acoustic resonance of the respiratory airways represents an optimal state for alveolar gas exchange. high frequency jet ventilation (HFJV) is a recognised form of mechanical ventilatory support that is used in both anaesthesia and critical care medicine. The technique differs from conventional modes of ventilatory support in both its relative tidal volume and respiratory rate. Several studies have shown that HFJV is capable of maintaining adequate gas exchange in cases where conventional methods have either failed or proved to be impractical. The main advantages of HFJV include lower peak and mean airway pressures, a reduction in pulmonary barotrauma and less disturbance to cardiovascular function.

The purpose of this case study is to present a development of the step-invariant approach to highpass digital filter design. It will be shown that the transfer function,G(z) obtained via the step invariant design method, may be made to approximate closely to the transfer function, G(z) obtained via the bilinear z-transform method. The analysis presented justifies this step-invariant design method, and shows that it is possible to convert a time domain (step-invariant) filter, G(z) to one that satisfies a frequency domain specification,G(z) . This is achieved by observing certain conditions and by employing a suitable gain term. The validity of the method is demonstrated using a practical example of a simple highpass filter and a digital phase-advance network.

The system described has shown that the use of a DSP in the role of measurement for control gives advantages over conventional systems. Increased speed, and the capability to use more complex algorithms leads to improved controller performance. Although designed specifically for a laboratory system, the necessity for improved measurement systems in industry is growing. Manufacturers are now starting to implement digital controllers, the performance of which will soon be limited by a lack of accurate information at high transfer rates. Parameters which were once deemed unmeasurable can now be constructed on-line and in real-time. This opens up greater possibilities for control system designers. Specialist hardware which was originally designed for communications processing will have an increasing role in control systems implementation due to the similarities in algorithm structure. Manufacturers have already seen this trend, and have begun to produce high performance devices specifically for these applications. It is now up to the systems engineer to recognise the advantages of the DSP for particular applications, and to use these to the full.

In this review a brief summary is given of the current trends and new directions in the development of algorithms, architectures and devices for signal processing. This, of necessity, is a personal view point, nevertheless, with significant new developments over the horizon, the subject area of signal processing is set to grow and grow. As a final statement, the integration of fast algorithms, parallel architectures, and high performance multiprocessors, the field of parallel signal processing and its applications is one which will lead to rich rewards.