tocAnalog Floating-Gate, On-Chip Auditory Sensing System InterfacesPaul Hasler, Senior Member, IEEE, Paul D. Smith, Member, IEEE, DFig. 1. Cooperative analog digital signal processing (CADSP) appI. A NALOG D IGITAL P ARTITIONINGA. Power and SpaceFig. 2. Guidelines on using ASP or DSP depending upon required rB. Signal-to-Noise Versus CostII. P ROGRAMMABLE A NALOG CMOS T ECHNOLOGYFig. 3. Typical circuit elements used in auditory signal processIII. S IGNAL -P ROCESSING C IRCUITSA. Frequency DecompositionB. Amplitude DetectionFig. 4. Our continuous-time noise suppression system. (a) The ovC. Weighted MultiplicationIV. N OISE S UPPRESSION FOR S PEECH E NHANCEMENTFig. 5. (a) Block diagram of a potential speech front-end systemV. A NALOG S IGNAL -P ROCESSING F RONT E ND FOR S PEECH R ECOGNIFig. 6. Cepstrum system output. The system input is a sequence oA. Continuous-Time CepstrumFig. 7. Basic circuit, architecture, and measurements from the VB. Continuous-Time VQFig. 8. Circuit design for our HMM branch element as well as theC. Continuous-Time HMMR. Ellis, H. Yoo, D. Graham, P. Hasler, and D. Anderson, A contiR. Sarpeshkar, Efficient precise computation with noisy componenC. Mead, Analog VLSI and Neural Systems . Reading, MA: Addison-WC. A. Mead, Neuromorphic electronic systems, Proc. IEEE, vol. 7P. Hasler and D. V. Anderson, Cooperative analog-digital signal P. D. Smith and P. Hasler, Analog speech recognition project, inP. Hasler and T. S. Lande, Overview of floating-gate devices, ciP. Hasler, C. Diorio, B. A. Minch, and C. A. Mead, Single transiP. Hasler, B. A. Minch, J. Dugger, and C. Diorio, Adaptive circuM. Kucic, P. Hasler, J. Dugger, and D. V. Anderson, ProgrammableT. S. Hall, P. Hasler, and D. V. Anderson, Field-programmable anM. Kucic, A. Low, P. Hasler, and J. Neff, A programmable continuP. Smith, M. Kucic, and P. Hasler, Accurate programming of analoD. Graham and P. Hasler, Capacitively-coupled current conveyer sH. Yoo, D. V. Anderson, and P. Hasler, Continuous time audio noiP. Smith, M. Kucic, R. Ellis, P. Hasler, and D. V. Anderson, CepP. Hasler, P. Smith, C. Duffy, C. Gordon, J. Dugger, and D. AndeJ. R. Deller, J. G. Proakis, and J. H. L. Hansen, Discrete-Time T. M. Massengill, D. M. Wilson, P. Hasler, and D. Graham, EmperiJ. Schurmann, Ed., Pattern Classification, A Unified View of StaT. Delbruck, Bump circuits for computing similarity and disimilaG. T. Tuttle, S. Fallahi, and A. A. Abidi, An 8 b CMOS vector A/G. Cauwenberghs and V. Pedroni, A low-power CMOS analog vector qP. Hasler, Continuous-time feedback in floating-gate mos circuitJ. Lazzaro, S. Ryckebusch, M. A. Mahowald, and C. A. Mead, WinneS. Ranals, N. Morgan, H. Bourlard, M. Cohen, and H. Franco, ConnP. Smith and P. Hasler, A programmable diffuser circuit based onJ. Lazzaro, J. Wawrzynek, and R. Lippmann, A micropower analog VP. Hasler, P. D. Smith, E. Farquhar, and D. V. Anderson, A neuroIEEE SENSORS JOURNAL, VOL. 5, NO. 5, OCTOBER 2005 1027Analog Floating-Gate, On-Chip AuditorySensing System InterfacesPaul Hasler, Senior Member, IEEE, Paul D. Smith, Member, IEEE, David Graham, Student Member, IEEE, Rich Ellis,and David V. Anderson, Senior Member, IEEEAbstract—This paper describes our current efforts towardcreating cooperative analog–digital signal-processing systems forauditory sensor and signal-processing applications. We addressresolution issues that affect the choice of signal-processing algo-rithms arriving from an analog sensor. We discuss current analogcircuit approaches toward the front-end signal processing by re-viewing major programmable analog building blocks and showinghow they can be interconnected to create a complete system. Wealso discuss our current IC approaches using this technology fornoise suppression, as well as our current analog signal-processingfront-end system for speech recognition. Experimental data ispresented from circuits fabricated using a 0.5m nwell CMOSprocess available through MOSIS.Index Terms—Analog cepstrum, analog hidden Markov model(HMM), analog signal processing (ASP), analog speech enhance-ment in noise, analog speech recognition, analog vector quantiza-tion (VQ), auditory signal processing, floating-gate circuits.NEW advances in analog very large-scale integration(VLSI) circuits have made it possible to perform opera-tions that more closely reflect those done in digital signal-pro-cessing (DSP) applications or that are desired in future DSPapplications. With these advances, analog circuits and systemscan beprogrammable, reconfigurable, adaptive, and at a densitycomparable to digital memories (for example, 100 000+ multi-pliers on a single chip). Therefore, with both DSP and analogsignal-processing (ASP) modalities feasible, more options arenow available when designing a signal-processing system.In this paper, we will discuss ASP in the context of severalaudio-processing systems. The comparable digital algorithmsare well understood and, since they are not novel, are not dis-cussed here. The purpose, then, of this paper is to demonstratethe analog options available when deciding where to partitionthe analog and digital parts of a system. First, we will addressresolution issues that affect the choice of signal-processing al-gorithms arriving from an analog sensor. Second, we will dis-cuss the building blocks of current analog circuit approachestoward front-end signal processing and the relationship to mod-eling biological cochleas. Third, we will discuss our current ICManuscript received November 8, 2002; revised September 27, 2003. Theassociate editor coordinating the review of this paper and approving it for pub-lication was Prof. P. M. Sarro.P. Hasler, D. Graham, and D. V. Anderson are with the Department of Elec-trical and Computer Engineering, Georgia Institute of Technology, Atlanta, GA30332-0250 USA (e-mail: [email protected]; [email protected];[email protected]).P. D. Smith was with the Department of Electrical and Computer Engineering,Georgia Institute of Technology, Atlanta, GA 30332-0250 USA. He is now withGtronix, Inc., Fremont, CA 94538 USA.R. Ellis is with Medtronic, Tempe, AZ 85281 USA.Digital Object Identifier 10.1109/JSEN.2005.854488Fig. 1. Cooperative analog–digital signal processing (CADSP) applied towardauditory sensor processing. We assume the typical model of signals comingfrom real-world sensors, which are analog in nature, that need to be utilizedby digital computers. Our approach is to perform some of the computationsusing