The ACT programmable transversal filter.
In the past 20 years, advanced digital integrated circuits have brought the power of digital signal processing to a variety of military and commercial electronic systems. However, in many systems in the areas of communications, radar, electronic warfare, image processing, test instrumentation and data storage, the speed of even the fastest digital signal processors falls far below what is needed. The common requirement in many of these systems is that signals with bandwidths of tens or even hundreds of megahertz must be processed in real time, often in a limited volume, with little power consumption. With the introduction of a new signal processing technology, called acoustic charge transport (ACT), the digital signal processing evolution may now be extended into many of these systems.
The ACT programmable transversal filter (PTF) is a sampled-analog signal processor with a digitally-programmable finite impulse response. This paper discusses the use of the ACT PTF as a wideband analog filter with user-programmable amplitude and phase response; some of the many applications that may utilize this capability include receiver IF filtering, interference cancellation and signal generation.
The fundamental enabling feature of ACT technology is the ability to implement on a single chip a wideband tapped delay line with programmable tap weights. The value of this function has long been known to engineers working in the adaptive signal processing field, where such an architecture is known as an adaptive linear combiner, adaptive finite impulse response (FIR) filter, or adaptive transversal filter.  A schematic diagram of a transversal filter is shown in Figure 1. In this figure, each block labelled T, often shown as [z.sup.-1] in signal processing texts, represents a unit delay element, and each block labelled h represents a weighting coefficient by which the delayed signals are multiplied prior to summation. The output of the transversal filter is given by,
[V.sub.out] = [summation] [h.sub.n] [V.sub.n]
where [V.sub.n] = the delayed input signal n = the number of delay and weighting elements in the filter
Thus, the output is seen to be the discrete convolution of the input signal with the impulse response of the filter. 
To understand how a linear combiner may be used as a filter, consider the case in which all the tap weights [h.sub.n] are initially set to +1. The output signal [V.sub.out] is then just the sum of n delayed versions of the input signal. Any frequency present in the input signal, for which the delay time T represents an integer number of hertz, will add constructively with all of the other delayed versions of the input signal. Conversely, any frequency component which undergoes another 0.5 Hz during the delay time T will reverse phase on every other tap and will, therefore, add destructively. Thus, the transversal filter passes certain frequency components of the input signal and blocks others. In the case of programmable transversal filters, the coefficients may be changed in order to vary the response of the filter. For example, to block the frequencies that were passed by the all one-filters (while passing those that were blocked), the user could choose to reverse the sign of the coefficient on every other tap. To reduce the filter sidelobes, the user may also elect to employ a weighting function that reduces the amplitude of the tap coefficients at both ends of the filter. It should be clear that the coefficients of the taps (the time-domain impulse response) determine the frequency-domain characteristics of the programmable transversal filter.
The ACT Programmable
ACT technology allows the implementation of a complete programmable transversal filter on a single GaAs chip. To understand the operation of the ACT device, consider the simplified ACT delay line, shown in Figure 2. In the ACT device, an input signal, shown as x(t), is applied to an ohmic input contact (IC) on a depleted n-type epitaxial layer on a semi-insulating GaAs substrate. On the same device, a high-Q transducer is used to generate a surface acoustic wave (SAW) that propagates across the device, from left to right, at 2864 m/s, the speed of sound in GaAs. Since GaAs is a piezoelectric material, the propagating surface wave induces within the epitaxial layer a series of electric fields that may be thought of as traveling potential wells. As each well passes under the input contact, a number of electrons proportional to the input voltage at that instant are injected into the well. The fields associated with the surface wave then transport these packets of electrons away from the input contact and along the ACT channel. Through this process, the analog input signal has been converted into a discrete-time signal (sampled in time, but not quantized in amplitude). This signal may be considered to be stored in the device during the transport process. It then may be sensed through a series of nondestructive sense (NDS) electrodes that measure but do not change the amount of charge in each packet. These electrodes form the taps that provide the delayed versions of the input signal required for transversal filtering. At the end of the ACT channel, an ohmic extraction contact (EX) is used to remove the charge from the wells and, if desired, to produce a sampled and delayed version of the input signal. 
The ACT tapped delay line has been used to implement a complete transversal filter through the addition of on-chip tap-weighting and memory circuits. As shown in Figure 3, the weighting of the taps is accomplished through the use of programmable attenuators that set the magnitude of the coefficient. The accumulation function required by the transversal filter is performed by two summing busses; these busses are connected to the inverting and the noninverting inputs of an offchip differential amplifier in order to permit bipolar tap weighting. The tap weights are stored in on-chip static random-access memory (SRAM). In order to change the response of the filter, the user simply supplies a data word containing the desired coefficient, the address of the tap to be loaded, and an enable signal at the time the change is desired.
The standard ACT programmable transversal filter has 128 taps that may be set to any of 31 values between +1 and -1 (5-bit tap weighting). The input sampling rate of the device is 360 MHz, and the tap spacing is 5.6 ns. This gives a single-tap bandwidth of 180 MHz and a filter Nyquist interval of 90 MHz. A single tap may be programmed in less than 1 [mu]s, and the entire device may be programmed in under 100 [mu]s. Details of the PTF architecture have been published previously. [4,5]
Example Applications of the PTF
Programmable IF Filtering
The ability to control the center frequency, bandwidth, amplitude and phase independently of the programmable transversal filter offers significant benefits to receiver designers. The variable center frequency of the filter allows the designer to utilize the optimum IF frequency (or even a variable IF frequency) to minimize the effects of mixing products and other unwanted signals. By varying the bandwidth of the ACT filter, the receiver bandwidth may be matched closely to the information bandwidth of the desired signal, minimizing the channel noise. Figure 4 shows a series of responses obtained by programming an ACT filter to vary its center frequency between 4 MHz and 84 MHz; the passband near 92 MHz is actually in the second Nyquist interval and is the result of the 84 MHz passband folding over the 88 MHz Nyquist frequency. The processing gain of the filter results in small insertion gain at low frequencies; a slight frequency roll-off in the individual tap response causes the insertion loss to increase to 5 dB at the upper end of the Nyquist interval. If identical response across the frequency range is required, the tap coefficients may be reduced for the low frequency filters to provide uniform processing gain.
In Figure 5, the filter bandwidth is varied from 2 MHz to 8 MHz in steps of 2 MHz. The filter sidelobe level remains approximately the same as the bandwidth is increased, but the amplitude within the passband decreases since wider-bandwidth responses use fewer taps, and therefore, have less processing gain. As in the variable-frequency case, uniform processing gain may be achieved by reducing the tap coefficients in the higher-gain filters.
The ability to vary the amplitude response of the ACT filter while maintaining linear phase is shown in Figure 6. The ACT filter also allows the user to vary the phase of the filter without changing the amplitude response, as shown in Figure 7. The value of independent control of filter amplitude and phase for the receiver designer is that variations in other portions of the system may be compensated in the ACT IF filter.
The precise gain and phase control of the ACT filter enable the implementation of a programmable notch filter that cancels narrowband interference in a wideband system. In this application, a feed-forward technique is known as invert-and-add is employed to cancel narrowband interference while not significantly degrading the desired wideband signal. As shown in Figure 8, the original input signal is applied both to the ACT filter and to a feed-forward (direct) path. The gain of the filter at the interference frequency is adjusted to equal precisely that of the direct path, and the delay of the filter is adjusted so that the signal passing through the filter is exactly 180[degrees] out of phase with the direct signal. The output of the ACT filter then is combined with the feed-forward signal; the phase-inverted signal produced in the filter is used to cancel the interference. In this way, frequencies outside the passband of the filter are not affected by the interference suppression.
The frequency response of the ACT interference canceller is known in Figure 9. In this figure, two responses are overlaid: the first shows the canceller programmed to produce a notch at approximately 33 MHz, and the second shows a notch at 57 MHz. In each case, notch depths of 35 to 40 dB are achieved. The increased amplitude response on each side of the notch is a characteristic of the invert-and-add approach; the width of the response depends on the relative delay between the feed-forward and filter path lengths. The frequency spectrum of a wideband pulse contaminated by a narrowband interferer is shown before and after cancellation in Figures 10a and 10b.
Since the output of a transversal filter is the convolution of the input signal and the filter's impulse response, the ACT PTF may be used to generate arbitrary signals simply by programming the appropriate tap coefficients and pulsing the device. For example, the increasing-frequency (up-chirp) waveform, shown in Figure 11, was produced by applying a narrow pulse to the input of an ACT filter in which the taps had been loaded with a Hanning-weighted cosine wave with a frequency sweeping from 5 MHz to 60 MHz. The same device may be used to produce a down-chirp waveform or other waveforms with different phases slopes or weighting functions. If the input to the device is changed from a short pulse to the inverse of the tap coefficients, the ACT device may be used as a matched filter for pulse compression applications.
The ability to perform a real-time correlation between a wideband input signal and a user-defined impulse response is useful in applications such as communications channel simulation and radar countermeasures. In communications channels, the variable delay and amplitude response of the ACT transversal filter may be used to simulate multipath by producing echoes that arrive hundreds of ns after the direct-path signal. As a radar training device or active countermeasure, the ACT device may be used to convolve a threat radar pulse with a user-defined radar cross-section profile to simulate the return from a complex target at any aspect angle. The small size of the ACT filter allows multiple devices to be used on a small platform to simulate effects such as glint and polarization rotation.
ACT technology has enabled the realization of a programmable FIR filter with variable center frequency, bandwidth, amplitude and phase. This device is now finding applications in communications, electronic warfare, image processing, test instrumentation and data storage systems. The next step in the development of this technology will be to sense the output of the ACT device and close the loop to the tap controller. Once this is achieved, truly adaptive signal processing functions such as chanell equalization, multipath cancellation, and automatic interference suppression may be performed.
Daniel A. Fleisch received his BS degree in physics from Georgetown University in 1974, and his MS and PhD degrees in space physics and astronomy from Rice University in 1977 and 1980, respectively. Presently, he is manager of Applications and Systems at Electronic Decisions Incorporated. He is responsible for developing applications for ACT technology in areas such as advanced communications, electronic countermeasures, test instrumentation, image processing, and data storage. Fleisch also oversees EDI's internal research and development programs. Prior to joining EDI in 1985, Fleisch was VP at Metratek Inc.
Glenn C. Pieters received his BSEE from the University of Illinois in 1983 and presently is completing his MSEE also at the Univeristy of Illinois. He has been involved in the research and development of ACT signal processors since 1981. Pieters has been an employee of Electronic Decisions Incorporated since its inception in 1985. He is presently manager of technical sales and service and has previously managed and participated in the development of several ACT processors, such as programmable transversal filters, selectable delay lines and analog memories. Pieters is a member of the IEEE.
(*) invited paper.
 Bernard Widrow and Samuel D. Stearns, Adaptive Signal Processing, Prentice Hall, 1985.
 Alan V. Oppenheim and Ronald W. Schafer, Discrete-Time Signal Processing, Prentice Hall, 1985.
 M. J. Hoskins and B. J. Hunsinger "Simple Theory of Buried Channel Acoustic Charge Transport in GaAs," J. Applied Phys., Vol. 55, No. 2, Jan. 1984, pp.413-426.
 R. W. Miller and R. J. Kansy, "Acoustic Charge Transport Digitally Programmable Transversal Filter Development," 1990 IEEE MTT-S Symposium Proceedings, Cat. #CH2848-0, Vol. 3, May 1990, pp 1111-1114.
 J. E. Bales, M. J. Hoskins and P. H. Sahm, "A GaAs ACT/IC Programmable Wide-Band Analog Signal Processor," 1990 IEEE GaAs IC Symposium, Cat. #CH2889-4, October 1990, pp. 23-26.
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||acoustic charge transport; analog-digital hybrid filter|
|Author:||Fleisch, Daniel A.; Pieters, Glenn C.|
|Date:||May 1, 1991|
|Previous Article:||GaAs W-band IMPATT diodes: the first step to higher frequencies.|
|Next Article:||Modeling GaAs MESFETs for intermodulation analysis.|