CEBE IAB Meeting, Sept 16-18, 2013 in Tallinn Research on Signal Processing Cooperation with ELIKO Competence Center in electronics and ICT by Mart Min mart.min@ttu.ee Thomas Johann Seebeck Department of Electronics Tallinn University of Technology 1 Signal Processing: what kind of? for what? Digital and analog processing (synthesis and analysis) for: 1. Synthesis and generation of excitation signals with predetermined bandwidth, waveform, spectral content and shape, for obtaining the most effective excitation for systems/substances to be measured, studied, tested. 2. Analysis of the response signals (the results of excitation) in: (a) - frequency domain; (b) - joint time-frequency domain; (c) - time domain, to obtain the maximal amount of information for identification of dynamic and predominantly time varying systems, circuits, materials, structures. Remarks: Our main identification method is impedance spectroscopy of both technical and living systems (also impedance spectro-tomography). Impedance – electrical (mostly), but also acoustical, optical, and mechanical. The terms bioimpedance and (electro-)chemical impedance mean also electrical impedance, but of biological or chemical matter. 2 Identification of dynamic systems is the goal Generation of Excitation Synthesis of Excitation Signals M u x ωexc and Texc waveform and energy Reference Processing of the Response Control . . . Impedance Ż(ω,t) Re{Z}; Im{Z} speed of changes . . . Control System under study D e m u x Processing of Response Signals ω and t domain: resolution, signal-to-noise, amount of info Ż(ω,t) Sampling Reference 3 Focus: finding the best excitation waveforms for the fast and wideband time dependent spectral analysis: intensity (Re & Im or M & φ) versus frequency ω and time t 4 Requirements to the Impedance Spectroscopy A. Fast measurement and signal processing in a wide frequency range; B. Simple architecture and electronic circuitry (simplicity, dependability); C. Low power (extremely low in some applications) and low voltage operation; Excitation waveform: a) easy to generate; b) easy to tune; c) covers the needed frequency range; d) generated energy must be concentrated into the BW of interest; e) effective energy packaging (low crest factor - less than 1.5); f ) simple processing of the response signal. Signal processing for performing deconvolution: a) simple algorithms, b) fast processing of the response signals, c) getting frequency domain but time dependent results – performing the joint time-frequency analysis. 5 Problems to be solved by using of chirps Impedance appears to be non-stationary - their spectra are time dependent. Examples: (a) cardiovascular system (beating heart, pulsating blood); (b) pulmonary system (breathing); (c) running bio-particles in a microfluidic device. Excitation must be: 1) as short as possible to avoid significant changes during the spectrum analysis; 2) as long as possible to enlarge the excitation energy (max signal-to-noise ratio). Which waveform is the best one? A unique property of chirp waveforms – scalability – enables to reach compromise between contradictory requirements (1) and (2) The questions to be answered: a. A chirp wave excitation contains typically hundreds and thousands of cycles. What could be the lowest number of cycles applicable if the fast changes take place? b. Are there any simpler rectangular waveforms (binary or ternary) to replace the sine wave based chirps in practical spectroscopy? 6 Scalable chirp signals: two chirplets 2 B. Scalability in time domain: duration Texc changes, BW = const = 100 kHz 1.0 0.8 12 cycles 0.5 48 cycles 0.2 0.0 -0.2 -0.5 -0.8 -1.0 0 100u 200u 300u 400u 500u Texc = 250 μs 600u 700u 800u 900u 1m Texc = 1000 μs 100m Bandwidth BW = 100 kHz = const 4.48 mV/Hz1/2 2.24 mV/Hz1/2 Energy E250μs = 125 V2∙μs Energy E1000μs = 500 V2∙μs Voltage Spectral Density @ 250μs = 2.24 mV/ 10m Hz1/2 1 mV / Hz1/2 1m Voltage Spectral Density @ 1000μs = 4.48 mV/ Hz1/2 100u Changes in the pulse duration Texc reflect in spectral density 10u BW = 100 kHz 1u 1k 10k 100k 1M 10M 7 Scalable chirp signals: two chirplets 1 A. Scalability in frequency domain: bandwidth BW changes, Texc = const = 250 μs 1.0 0.8 48 cycles 12 cycles 0.5 t 0.2 0.0 -0.2 -0.5 -0.8 -1.0 0 25u 50u 75u 100u 125u 150u 175u Texc = 250 μs 200u 225u 250u Texc = 1000 μs 100m Excitation time Texc = 250 μs = const Excitation energy Eexc = 0.5V2 ∙250 μs = 125 2.24 mV/Hz1/2 V2∙μs 10m 1 mV / Hz1/2 1.12 mV/Hz1/2 Voltage Spectral Density @ 100 kHz = 2.24 mV/ Hz1/2 BW = 100 kHz 1m Voltage Spectral Density @ 400 kHz = 1.12 mV/Hz1/2 BW = 400 kHz 100u Changes in the frequency span BW reflect in spectral density 10u 1u 1k 10k 100k 1M 10M 8 A very short Chirplet - Half-cycle linear sin θ RMS spectral density (relative) sin θ(t) 10 10 1 -40 dB/dec 1 θi= θ(ti) θfin = π θ 1 θ0= 0 cos θ 0 t =0 t ti tfin= Tch = 10 μs 1 2.26 mV/ 10 100m -1 f 10m ffin= 100 kHz 10-2 100kHz f(ti) f0= 0 t=0 t ti t 10-3 1m tfin= Tch = 10 μs 100u 10-4 Texc = Tch = 10 μs, BW = 100 kHz Instant frequency Hz1/2 1k 1k 10k 10k 100k 100k d (t ) 2 f fin t / T,ch, rad/s - a linear frequency growth dt 1M 1M f, 10M Hz Current phase t (t ) dt 2 f fin t 2 / 2Tch , rad; Generated chirplet sin t sin 2 ( t ) dt 2 f t / 2Tch fin 9 A very short Chirp - 2x quarter-cycle linear chirplet Normalised level RMS spectral density, normalised 1.0 20 0.5 0.0 0 Time, μs -0.5 -1.0 0 2 4 6 8 10 12 14 16 18 20 Frequency, MHz (max 100kHz) 0.10 0.08 f =fmax(t /Tch 0.06 )2 0.04 -80 dB/decc -20 -40 -60 -80 1k 0.02 0.00 0 2 4 6 8 100k 1M 10M Frequency, Hz 10 12 14 16 18 20 Time, μs 10k C 2t sin 2 kch t 3 / 3 10 Spectra and power of binary/ternary chirps 0 Binary (0) 18 100kHz 30 Ternary (30) Pexc – excitation power within (BW)exc=100kHz Binary(0): Pexc= 0.85P Ternary (21.2): Pexc= 0.94P – max. possible! 11 Classical sinc waveform – mathematically the best Relative time 12 Several sine waves simultaneously – Multisine excitation Fast simultaneous measurement at the specific frequencies of interest! + Simultaneous/parallel measurement and analysis (fast); + Frequencies can be chosen freely; +/- Signal-to-noise level is acceptable; − complicated synthesis restricts the number of different frequency components. Signal space is limited between +1 and -1 (ΣAi = 1) Max crest factor maxCF = ΣAi / (RMS)Σ = 2.83 Min(RMS)Σ = 0.36 (worst case) 0 Max(RMS)Σ = 0.72 (optimised phases) 13 Crest factors CF of optimised multisine excitation (a sum of n sine wave components, n = 3 to 20) CF = ΣAi /RMS for optimally synthesized multisine signals The best known before For a single sine wave CF=√2=1.414 Jaan Ojarand’s algorithm 14 Optimised multisine waveform Relative time 15 Binary multifrequency waveform Relative time Less than 10% of total RMS 16 Synthesized multifrequency binary sequences (4 components – 1, 3, 5, 7f) Equal-level components Growing-level components ! 17 Energy and RMS of different excitation waveforms BMF- binary multifrequency MS- multisine A single sine wave has: energy- 50%, RMS - 71% (less than MS!) bipolar sinc sinc 1- binary multifrequency (BMF) 2- optimal multisine (MS) 3- modified sinc (bipolar) 5- sinc (classic) 18 Collaboration with industry through ELIKO Impedance spectroscopy devices using MBS: laboratory devices prototyped in ELIKO The Main Unit The Pick-up Unit I+ Clock & Sync USB excitation Differential Current Source DSP unit PGA1 +5V Power Supply Unit response Voltage Buffer Cable 1 I- Ż(jω) . VZ Cable 2 20 Texas The project with Electrolux Italy S.p.a Partners: Food and Fermentation Competence Center and ELIKO Meat quality assessment CAROMETEC A/S just bought a license to use the impedance spectroscopy method (CEBE patent) for meat quality assessment (13.10.2013). Carometec is a world leader in production of meat quality equipment for the food industry 22 Real-time in vivo identification of various physiological condition of organs 23 using a range of needles. The foundations are: the different electrical properties of human tissues (bioimpedance), advanced measurement technology (CEBE patent) we gave over to Injeq Oy, Finland, and proprietary needle designs (Injeq’s patent) Summary The research center CEBE is founded for making fundamental science. The scientific results can be and have been transferred into industry and commercialised using Technology Competence Centres as ELIKO – electronics and ICT, and FFCC – food and fermentation. Thank you for listening! 24