Sunday, October 6 9:00 am – 12:00 noon
SESSION E: SIGNAL PROCESSING, PART 3

Chair: Jayant Datta, Discrete Labs,

E-1 Computationally Efficient Inversion of Mixed Phase Plants with IIR Filters—Timoleon Papadopoulos, Philip A. Nelson, University of Southampton, Southampton, UK

Inverse filtering in a single or in multiple channels arises as a problem in a number of applications in the areas of communications, active control, sound reproduction, and virtual acoustic imaging. In the single-channel case, when the plant C(z-1) sought to be inverted has zeros outside the unit circle in the z-plane, an approximation to the inverse 1/C(z-1) can be realized with an FIR filter if an appropriate amount of modeling delay is introduced to the system. But the closer the zeros of C(z-1) are to the unit circle (either inside or outside it), the longer the FIR inverse has to be, typically several tens of times longer than the plant. An off-line implementation utilizing a variant of the backward-in-time filtering technique usually associated with zero-phase FIR filtering is presented. This forms the basis on which a single-channel mixed phase plant can be inverted with an IIR filter of order roughly double than that of C(z-1), thus decimating the processing time required for the inverse filtering computation.
Convention Paper 5659

E-2 Optimal Filter Partition for Efficient Convolution with Short Input/Output Delay—Guillermo García, Creative Advanced Technology Center, Scotts Valley, CA, USA

A new algorithm to find an optimal filter partition for efficient long convolution with low input/output delay is presented. For a specified input/output delay and filter length, our algorithm finds the non-uniform filter partition that minimizes computational cost of the convolution. We perform a detailed cost analysis of different block convolution schemes and show that our optimal-partition finder algorithm allows for significant performance improvement. Furthermore, when several long convolutions are computed in parallel and their outputs are mixed down (as is the case in multiple-source 3-D audio rendering), the algorithm finds an optimal partition (common to all channels) that allows for further performance optimization.
Convention Paper 5660

E-3 Filter Morphing—Topologies, Signals and Sampling Rates—Rob Clark^{1, 2}, Emmanuel Ifeachor², Glenn Rogers¹ - ¹Allen & Heath Limited, Penryn, Cornwall, UK; ²University of Plymouth, Plymouth, UK

Digital filter morphing techniques exist to reduce audible transient distortion during filter frequency response change. However, such distortions are heavily dependent on signal content, frequency response settings, filter topology, interpolation scheme, and sampling rates. This paper presents an investigation into these issues, implementing various filter topologies using different input stimuli and filter state change scenarios. The paper identifies the mechanisms causing these distortions, specifying worst case filter state change scenarios. The effects of existing interpolator schemes, finite word length, and system sampling rates on signal distortion are presented. The paper provides an understanding of filter state change, critical in the design of filter morphing algorithms.
Convention Paper 5661

E-4 Evaluation of Inverse Filtering Techniques for Room/Speaker Equalization—Scott G. Norcross, Gilbert A. Soulodre, Michel C. Lavoie, Communications Research Centre, Ottawa, Ontario, Canada

Inverse filtering has been proposed for numerous applications in audio and telecommunications, such as speaker equalization, virtual source creation, and room deconvolution. When an impulse response (IR) is at non-minimum phase, its corresponding inverse can produce artifacts that become distinctly audible. These artifacts produced by the inverse filtering can actually degrade the signal rather than improve it. The severity of these artifacts is affected by the characteristics of the filter and the method (time or frequency domain) used to compute its inverse. In this paper objective and subjective tests were conducted to investigate and highlight the potential limitations associated with several different inverse-filtering techniques. The subjective tests were conducted in compliance with the ITU-R MUSHRA method.
Convention Paper 5662

E-5 Using Subband Filters to Reduce the Complexity of Real-Time Signal Processing—J. Michael Peterson, Chris Kyriakakis, University of Southern California, Los Angeles, CA, USA

Several high quality audio applications require the use of long finite impulse response (FIR) filters to model the acoustical properties of a room. Various structures for sub-band filtering are examined. These structures have the ability to divide long FIR filters into smaller FIR filters that are easier to use. Two structures will be discussed to process the signals in a real-time manner, time-convolution of spectrograms, and generalized filter banks. Also filter estimation will be discussed.
Convention Paper 5663

E-6 Noise Shaping in Digital Test-Signal Generation—Stanley P. Lipshitz¹, John Vanderkooy¹, Edward V. Semyonov² - ¹University of Waterloo, Waterloo, Ontario, Canada; ²Tomsk State University of Control Systems and Radioelectronics, Tomsk, Russia

In an earlier paper we put forth an idea to use noise-shaping techniques in the generation of digital test signals. The previous paper proposed using noise shaping around an undithered quantizer to generate sinusoidal digital test signals with spectra having error nulls at the harmonics of the signal frequency, thus making digital distortion measurements of very great dynamic range possible. We extend this idea in this present paper in a number of ways. We show a) that dither is necessary in order to suppress spurious artifacts caused by the nonlinearity of an undithered noise shaper; b) that wider and deeper nulls at the harmonic frequencies can be achieved by using higher-order noise-shaper designs; c) that IIR filter designs can moderate the increased noise power that accompanies an increased FIR filter order; and d) some other novel uses of noise shaping in digital signal generation.
Convention Paper 5664