BLIND CORRECTION OF GAIN AND TIMING MISMATCHES FOR A TWO-CHANNEL TIME-INTERLEAVED ANALOG-TO-DIGITAL CONVERTER 1 inch 1cm Munkyo Seo, Mark Rodwell, Upamanyu Madhow Department of ECE, University of California, Santa Barbara, CA 93106 Email: {mkseo, rodwell, madhow}@ece.ucsb.edu 1cm analog input digital output ADC 1 How to MEASURE Mismatches? Challenge: Mismatches Why Time-Interleaving? 1 Samples@ t= 1, 2, 3,… 2 Samples@ t=0.8,1.8,2.8,… 1 Gain= 1.0 2 Gain= 1.2 (1) Training method (off-line) G=? ∆t=? ADC 2 Training signals 1 2 1 2 1 2 1 2 Sampling Rate = Nchannel (Unit Sampling Rate) (2) Blind method (online) Analog Circuitry 1 2 Drift, process variation 1 1 2 2 (2) Digital post-correction 1 No interruption System interruption Calibration detuning Good accuracy Mismatches creates distortion. Must correct these!! (1) Analog pre-correction ?? Unknown input ?? 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 How to CORRECT Mismatches? Digital Filter 2 Tracks time-varying error Reliable, flexible, scalable Compromised accuracy 1cm 1 2/1 2 (∆G0, ∆t0) 2/1 Parameterized Filter Bank (∆G, ∆t) (∆G0, ∆t0) Find (∆G, ∆t) such that output auto-correlation is shift-independent. It can be shown, NO FALSE CORRECTION (i.e., such (∆G, ∆t) is unique) 10 10 10 10 10 10 10 10 10 10 ∆G= ±10% 14-b A/D boards Signal ∆t= ±5% Nsample=102~106 100 MSPS ADC chip Filter Summary Logic Blind Mismatch analyzer Correction √ Clock 0 Filter G (uncorrect ed) -1 ~ G G , ' WIDE' -2 -3 ~ G G , ' SINE' -4 -5 10 10 Experimental Result 2 10 3 0 10 4 10 Nsample 5 10 6 Amplitude (dB) Normalization 10 Standard deviation Assume Wide-Sense Stationary (WSS) input. Let’s restore output WSS 30 Monte-Carlo runs Input: ‘SINE’, and ‘WIDE’ Standard deviation Mismatch yields “uneven” output Simulation Result G (uncorrect ed) -1 ~ -3 ~ G G , ' SINE' -4 -20 -20 -40 _X O 2 4 6 Z -100 -120 0 -40 20 40 5 7 3 -80 Y 8 60 -100 80 100 -120 0 2 10 3 10 Nsample 4 10 √ Proposed blind method Generality Unified framework _X 2 4 6 Z 20 40 5 7 3 Y 8 60 O 80 100 √ First Uniqueness proof (for M=2) √ Simulation / Experiment 80 60 1 2 1 2 1 2 40 -5 10 2 -60 -60 -80 0 After cal. After Blind Calibration 1 100 G G , ' WIDE' -2 0 Before cal. Before Blind Calibration 1 1 SFDR SFDR (dB) Proposed Blind Algorithm 1cm 5 10 6 20 0 0.1 0.2 0.3 0.4 0.5 WSS assumption invalid Algorithm fails