Jim Schroeder

Adaptive mean/median filtering

The use of median and averaging filters is fairly routine in signal processing applications. One problem in using such algorithms is the lack of objective criteria by which to decide whether an averager or a median filter is more appropriate. We formulate an L/sub p/ (1/spl les/p/spl les/2) normed filter where p is chosen as a function of the kurtosis of the residual vector; we restrict attention in this work to a mean filter (p=2) and a median filter (p=1). In order to highlight the effectiveness of this filtering algorithm we demonstrate reduced sum squared error by adaptively filtering a sinusoid in the presence of both additive white Gaussian noise and an impulsive noise component.The use of median and averaging filters is fairly routine in signal processing applications. One problem in using such algorithms is the lack of objective criteria by which to decide whether an averager or a median filter is more appropriate. We formulate an L/sub p/ (1/spl les/p/spl les/2) normed filter where p is chosen as a function of the kurtosis of the residual vector; we restrict attention in this work to a mean filter (p=2) and a median filter (p=1). In order to highlight the effectiveness of this filtering algorithm we demonstrate reduced sum squared error by adaptively filtering a sinusoid and a test image in the presence of both additive white Gaussian noise and an impulsive noise component.

Model-based filter design by minimizing median of square of residuals

Model-based digital filter design may be an attractive technique if the desired impulse response, perhaps measured in the field, closely matches a simple time series model, such as an autoregressive model. For autoregressive model based filters, a least squares solution is convenient for computational reasons, but is adversely affected by data outliers, such as a severe noise spike. Previously, the authors have shown that an Lp(p=1) may generate a robust solution in certain cases, however, such an estimator, although more robust than least squares methods, suffers breakdown when the data outliers are too frequent or occur at end points of the data record. The present paper demonstrates the increased robustness of a model based filter design via choosing model coefficients by minimizing the median of the square of the residuals. >

Robust FSK sinusoidal frequency estimation

It a relatively short frequency shift keyed (FSK) sinusoidal burst signal is available for analysis, transient effects within the analysis window may be detrimental to linear predictive model based frequency estimation methods such as the forward-backward linear predictive technique. Signal transients are manifested as spikes within the residual vector produced from a linear predictive based approach to frequency estimation. For such cases, it is demonstrated that a robust estimator, such as an estimator generated via an L/sub 1/ normed error criterion, can produce less biased frequency estimates. Additionally, it is shown that sinusoidal frequency estimates generated via an L/sub 1/ normed solution are insensitive to initial signal phase in contrast to least squares based estimates. It is demonstrated that a robust Wigner distribution can be estimated based upon the ideas presented in this paper.<<ETX>>

Simulation of RBDS AM subcarrier modulation techniques to determine BER and audio quality

This paper describes the modeling and simulation of candidate AM subcarrier systems designed to transmit digital information for display on the dial of next generation radios. AM subcarrier technology still remains under-developed and its possible applications untapped. One such application of AM subcarriers is the Advanced Traffic/Traveler Information System. In the proposed system, the digital signal is inserted between the AM carrier and the audio sidebands. This paper investigates the effect of mutual interference between the audio and digital programs for a variety of digital modulation schemes, bit rates, and channel conditions. The GMSK subcarrier model exhibited the best overall performance in terms of spectrum efficiency and output signal-to-noise ratio.

Model based filter design by minimizing median of square of residuals

Model-based digital filter design may be an attractive technique if the desired impulse response, perhaps measured in the field, closely matches a simple time series model, such as an autoregressive model. For autoregressive model based filters, a least squares solution is convenient for computational reasons, but is adversely affected by data outliers, such as a severe noise spike. Previously, the authors have shown that an Lp(p=1) may generate a robust solution in certain cases, however, such an estimator, although more robust than least squares methods, suffers breakdown when the data outliers are too frequent or occur at end points of the data record. The present paper demonstrates the increased robustness of a model based filter design via choosing model coefficients by minimizing the median of the square of the residuals.<<ETX>>Model based digital filter design may be an attractive technique if the desired impulse response, perhaps measured in the field, closely matches a simple time series model, such as an autoregressive model. For autoregressive model based filters, a least squares solution is convenient for computational reasons, but is adversely affected by data outliers, such as a severe noise spike. Previously, we have shown that an Lp (p=1) may generate a robust solution in certain cases, however, such an estimates, although more robust than least squares methods, suffers breakdown when the data outliers are too frequent or occur at end points of the data record. We demonstrate the increased robustness of a model based filter design via choosing model coefficients by minimizing the median of the square of the residuals.<<ETX>>

Burst sinusoidal frequency estimation with short record length via L/sub 1/ normed linear predictive modeling

The authors demonstrates that estimating the two sinusoidal frequencies of a burst sinusoidal frequency shift keyed (FSK) waveform via a least squares based linear predictive algorithm may result in significant frequency bias even in the absence of additional noise. Transient effects present from uncertain burst signal location within the analysis window may be detrimental to least squares based frequency estimation methods. Such signal transients are manifested as spikes within the residual vector produced from a linear predictive model based approach to frequency estimation. For such cases, it is demonstrated that a robust estimator, such as an estimator generated via a L/sub 1/ normed error criteria, may produce significantly less biased frequency estimates. Additionally, it is shown that sinusoidal frequency estimates generated via a L/sub 1/ normed solution are insensitive to initial signal phase in contrast to least squares based estimates.<<ETX>>

Current results in fast algorithm computational complexity analysis, computer architecture design, and VLSI hardware advances with applications to digital signal processing

As postulated in the editorial for this issue, quantum improvements in signal processing capability require advancement within the fields of algorithm design, computer architectures, and hardware capability. Ideally, algorithm design advances concurrently and synergistically with at least one other component, for example, computer architecture. Naturally, such an observation is hardly new: Quoting from a 1958 reference [l], “A general purpose digital computer can, in principle, solve any well defined problem . . . However, they are relatively inept at solving many problems where the data is arranged naturally in spatial form. It appears that efficient handling of problems of the type mentioned above cannot be accomplished without some form of parallel action.” Thus, in this example which predates hardware advances necessary for successful implementation, we see an early realization that computer architecture (parallel computation) must be coupled to algorithm design (two-dimensional signal processing). Required hardware is now available for constructing a computational engine that allows the simultaneous design of digital signal processing algorithms and a corresponding suitably matched computer architecture. It is unreasonable to expect a single researcher or small team of researchers to achieve advancement along all three, or even two, fronts. In recognition of this, major conferences within the signal processing community, such as the International Conference on Acoustics, Speech, and Signal Processing (ICASSP) and the European Signal Processing Conference (EUSIPCO), bring researchers from all three disciplines together at regular intervals in order to facilitate information transfer between these commonly disjointed efforts and publish technical papers in a common proceedings. In this spirit, for this overview we highlight some key results from ICASSP ‘90, April 3-6, 1990, held in Albuquerque, New Mexico. ICASSP ‘90 consisted of 60 separate technical sessions, with 752 papers presented, but to keep this task manageable we review just five technical sessions: VLSI Algorithms/Architectures, Transforms, VLSI Signal Processors, Computational Complexity and Fast Algorithms, and VLSI Array Signal Processing. Although this sampling represents much less than 10% (we necessarily must omit numerous high-quality papers) of all papers presented at ICASSP ‘90, we feel this is sufficient to present a flavor of current research efforts in the areas of algorithm computational complexity, computer architectures tailored to signal processing requirements, and VLSI hardware advances. It is noted that we do not review all papers in relevant sessions due to space limitations. The papers selected are chosen for purposes of representation rather than technical merit; we reserve judgment of research quality for the exclusive domain of the technically sophisticated ICASSP proceedings readership. The author apologizes in advance for omitting mention of many very fine research papers from ICASSP’SO. Additionally, somewhat arbitrarily, we reserve discussion of neural network-based algorithm research and optical signal processing research for future issues of the newsletter. Here, we hold our focus on “classical” algorithmic approaches that combine at least two of the three computational axes (generally a combination of the algorithm axis with the architecture axis) presented in the Editorial introducing this newsletter. Against these qualifications we

Suboptimal robust sinusoidal frequency estimation

It has been shown that an L/sub p/ normed solution to linear prediction equations may be advantageous in certain time-series modeling applications. Specifically, in the context of sinusoidal frequency estimation from data contaminated by impulsive noise, choosing p=1 may result in less biased frequency estimates. Computing L/sub p/ normed solutions requires iterative methods that are computationally expensive. By plotting prediction error filter root migration in the complex z-plane as a function of iteration step one is able to construct a suboptimal weighted least-squares solution. This solution, though not an L/sub p/ normed solution, is still robust against impulsive noise and is not as computationally expensive as iterating to a true L/sub p/ normed solution.<<ETX>>

Robust spectral estimation in the presence of muitipath

Previous research has shown that increased frequency resolution may be possible in the presence of impulsive noise utilizing an I.p (p = 1) solution to forward backward linear prediction equations (FBLP) . This work considers the problem of accurately estimating the frequency of a single sinusoid in the presence of a multipath component plus additive white Gaussian noise. It is shown that when a least squares solution to theFBLPemations resultsin fremencvbias. an G- 7; =- 1) normed solution mgy elTminate or significantly reduce the bias. Lntroductian spectral estimation has applications to many fields such as radar. sonar. seismic I