A 1 V 92 dB SNDR 10 kHz Bandwidth Second-Order Asynchronous Delta-Sigma Modulator for Biomedical Signal Processing

senso s Le e A1V92dBSNDR 10 kHz Bandwid h Second-O de Asynch onous Del a-Sigma Modula o o Biomedical Signal P ocessing Vilém Kled owe z * , Lukáš Fujcik , Roman P okop and Jiˇ í Háze Depa men o Mic oelec onics, B no Uni e si y o Technology (BUT), Technická 3058/10, 61600 B no, Czech Republic; ujcik@ u b .cz (L.F.); p okop @ u b .cz (R.P.); haze@ u b .cz (J.H.) *Co espondence: kled owe z@ u b .cz; Tel.: +420-541-146-101 Recei ed: 30 June 2020; Accep ed: 23 July 2020; Published: 25 July 2020   Abs ac : In his pape , a second-o de asynch onous del a-sigma modula o (ADSM) is p oposed based on he ac i e-RCin eg a o s. The ADSM is implemen ed in he 0.18 µ m CMOS Logic o Mixed-Signal/RF, Gene al Pu pose p ocess om he Taiwan Semiconduc o Manu ac u ing Company wi h a cen e equency o 848 kHz a a supply ol age o 1 V wi h a 92 dB peak signal- o-noise and dis o ion a io ( SNDR ), which co esponds o 15 bi esolu ion. These pa ame e s we e achie ed in all he endogenous bioelec ic signals bandwid h o 10 kHz. The ADSM dissipa ed 295 µ W and had an a ea o 0.54 mm 2 . The p oposed ADSM wi h a high esolu ion, wide bandwid h, and ail- o- ail inpu ol age ange p o ides he uni e sal solu ion o endogenous bioelec ic signal p ocessing. Keywo ds: asynch onous del a-sigma modula o (ADSM); cen e equency; ope a ional ampli ie ; biomedical signals; biosenso s 1. In oduc ion Biomedical elec onics ha e acqui ed signi ican a en ion in heal hca e, wi h a ocus on he de elopmen o biosenso s ha enable online moni o ing, de ec ion, p e en ion, and pe sonalized medicine o a a ie y o ch onic and acu e diseases. Especially in he las ew yea s, he e has been g owing in e es in he design o biomedical wi eless senso s [ 1 – 3 ]. Biomedical signals can be subdi ided in o wo majo classes: (1) endogenous signals ha a ise om na u al physiological p ocesses and a e measu ed wi hin o on li ing c ea u es (e.g., (EOG), elec oencephalog am (EEG), elec oca diog am (ECG o EKG), elec omyog am (EMG), empe a u e, blood glucose, e c.) and (2) exogenous signals applied om he ou side (gene ally nonin asi ely) o measu e in e nal s uc u es and pa ame e s. Endogenous bioelec ic signals a e in a iably small, anging om single mic o ol s o o e 100 mV. Thei bandwid hs ange om DC o pe haps 10 kHz a mos [ 4 – 7 ]. The ol age and equency anges o some common biopo en ial signals a e shown in Figu e 1. A gene al biomedical sys em consis s o an ene gy sou ce, a di e en ial ampli ie , analog- o-digi al con e sion (ADC), digi al signal p ep ocessing, and a communica ion subsys em. The ADC is one o he key building blocks, which enables con e ing analog signals om biomedical senso s o a digi al o ma ha can be easily p ocessed and analyzed. Fo he design o he ADC, many au ho s choose he SARa chi ec u e due o i s sui abili y o low-powe and low- ol age equi emen s [ 8 – 11 ]. Recen ly, del a-sigma ( ∆Σ ) ADC has been gaining mo e and mo e popula i y. Compa ed o o he con e sion echniques, ∆Σ ADCs co e he wides con e sion egion o he esolu ion- e sus-bandwid h plane, p o iding he mos e icien solu ion o digi ize di e se ypes o signals in many di e en applica ions such as biomedical ones [ 12 ]. The e exis wo basic ypes o ∆Σ modula o s: disc e e- ime (DTDSM) and con inuous- ime (CTDSM). The DTDSM is mo e Senso s 2020,20, 4137; doi:10.3390/s20154137 www.mdpi.com/jou nal/senso s Senso s 2020,20, 4137 2 o 13 a ac i e o high- esolu ion applica ions due o i s highe linea i y and accu acy. On he o he hand , less s ingen ampli ie speed speci ica ions a e equi ed in CTDSM due o he absence o swi ches in he ac i e-RC in eg a o , allowing achie ing a highe speed o ope a ion and lowe powe consump ion. The asynch onous ∆Σ modula o (ADSM) can be conside ed as a special ype o CTDSM. ADSM is simple , does no equi e any clocking, ma ches well wi h mains eam CMOS echnology, and can ope a e a low cu en and supply ol ages [ 13 – 15 ]. A compa ison be ween DTDSM, CTDSM, and ADSM is shown in Table 1. 10−1 100101102103104 10−5 10−4 10−3 10−2 10−1 F equency (Hz) Vol age (V) EOG (elec ooculog am) EEG (elec oencephalog am) EEG (elec oca diog am) DC po en ial EMG (elec omyog am) AAP (axon ac ion po en ial) Figu e 1. Vol age and equency anges o some common biopo en ial signals. Table 1. Compa ison be ween DTDSM, CTDSM, and ADSM. DTDSM CTDSM ADSM + Synch onous sys em + Synch onous sys em + Immuni y o clock ji e + High esolu ion + Implici an ialiasing il e + Implici an ialiasing il e + Highly linea SC in eg a o + Highe sampling equency + Simple ci cui + Accu a ely de ined in eg a o + Relaxed ope a ional ampli ie + Relaxed OpAmp gains and ans e unc ion speed equi emen s speed equi emen s + Low sensi i i y o clock ji e + Highe con e sion speed + Highe con e sion speed and excess loop delay + Low powe + Low powe + Low sensi i i y o DAC se ling ime + Do no equi e a clock - Low con e sion speed - Sensi i i y o clock ji e - Complex decoding scheme - Requi ed p e-an ialiasing il e - Excess loop delay - Lack o noise shaping - Requi ed non-o e lapping - Lowe esolu ion - Lowe esolu ion clock gene a o In se e al publica ions, DTDSMs a e used o biomedical signal p ocessing [ 1 , 16 , 17 ]. The e also exis s solu ions u ilizing ADSM [ 18 – 20 ]. These ADSM a e dis inguished by e y low powe consump ion in he o de o ens o nanowa s. Howe e , hei bandwid h is e y low in he o de o ens o He z. The p oposed ADSM co e s he ull bandwid h o endogenous bioelec ic signals up o 10 kHz. The di e en ial inpu ange equals VDDA wi h a 0.5 V e e ence le el ( VCM ). The p oposed ADSM wi h high esolu ion, wide bandwid h, and ail- o- ail inpu ol age ange p o ides he uni e sal solu ion o endogenous bioelec ic signal p ocessing. The ci cui no only o e s an al e na i e o he de eloped CTDSMs and DTDSMs, bu i also ills he gap be ween published ADSMs, which do no allow p ocessing he ull spec um o biomedical signals acco ding o Figu e 1excep o hose wi h a e y high bandwid h in he o de o MHz. An impo an pa ame e o ADSM is he cen e equency, he calcula ion o which is pa o his wo k. The ollowing sec ions p o ide he de ails o ou app oach. Senso s 2020,20, 4137 3 o 13 2. Asynch onous Del a-Sigma Modula o The e a e wo majo ypes o a chi ec u e o ∆Σ modula o s. The i s one is he single-loop and he second he mul i-loop a chi ec u e. Mul i-loop a chi ec u es a e commonly deno ed as cascade o MASH (mul i-s age noise shaping). A majo d awback o MASH modula o s is ha p ecise ma ching o he analog and digi al signal p ocessing pa hs is equi ed o a oid la ge e o s (quan iza ion noise leakage) caused by in eg a o gain coe icien a ia ions. Because RC in eg a o s a e used in his design, whe e a ia ions o abou 20% in he RC ime cons an can be expec ed, he single-loop a chi ec u e was chosen in his wo k. I s s onge abili y o achie e high SNDR since i does no su e om ma ching e o s, which se e ely a ec MASH modula o s, is he majo ad an age in he design. The block diag am o he second-o de ADSM based on he cascade o in eg a o s wi h dis ibu ed eedback (CIFB) opology is shown in Figu e 2. C1 R1 VCM R2 C2 R3 VCM R4 VCM VIN( ) I1( ) VOUTP( ) +VREF -VREF VOUTN( ) VY1( ) VY2( ) VDDA VSSA VDDA VSSA I3( ) I4( ) I2( ) Figu e 2. Simpli ied schema ic o he second-o de ADSM wi h he CIFB opology. The ci cui consis s o wo in eg a o s and a bina y quan ize wi h hys e esis. The ou pu VOUTP (o VOUTN ) is a pulse wid h modula ed squa e wa e o pe iod TPER wi h a pulse wid h TPW . The du y cycle d is p opo ional o he ampli ude o he inpu signal (Equa ion (1)). Mo eo e , he pe iod TPER o he asynch onous modula o ou pu signal is modula ed by he no malized inpu ol age VIN (Equa ion (2)) [21]. d=VIN +1 2=TPW TPER (1) and: 0 c =1−V2 IN and | |<1 (2) whe e 0 is he ou pu ca ie equency, c is he maximum alue o 0 , namely he cen e equency, and |VIN|<1 is he no malized inpu ampli ude. The cen e equency o ADSMs de e mines he ca ie - o-bandwid h a io ( CBR = c/( 2 B) , whe e B is he inpu signal bandwid h), which is he a io be ween he cen e equency and he signal bandwid h. This a io is equal o he o e sampling a io ( OSR ) in synch onous del a-sigma modula o s. I de e mines he minimal cen e equency equi ed o a ce ain con e sion accu acy. The c i ical condi ion can occu when VIN is close o he ull scale. The ou pu equency will dec ease, and he high- equency dis o ions a ound he cen e equency shi o he low- equency egion. Consequen ly, dis o ions can leak in o he baseband and ad e sely a ec he modula o linea i y o la ge inpu ampli udes. The e o e, he cen e equency should be se a away om he baseband o a oid hese componen s shi ing in o he signal baseband, and a high o de il e is equi ed o a enua e hese ou -band componen s. In o de o achie e a high cen e equency wi hou degene a ion o he linea i y, he second-o de opology was chosen. To calcula e he cen e equency o he p oposed ADSM, he in eg a o s’ ou pu ol ages a e exp essed as: VY1( ) = −I1( ) C1 +I2( ) C1 +VCM (3) Senso s 2020,20, 4137 4 o 13 VY2( ) = −I3( ) C2 +I4( ) C2 +VCM (4) whe e I1( ),I2( ),I3( ), and I4( )can be exp essed as: I1( ) = VIN( )−VCM R1 (5) I2( ) = VREF( )−VCM R2 (6) I3( ) = VY1( )−VCM R3 (7) I4( ) = VREF( )−VCM R4 (8) In o de o ind he cen e equency, he iming diag am in Figu e 3is conside ed, which co esponds o he schema ic in Figu e 2. VIN VDDA VSSA VCM VY1 VY1(mean) ∆VY1 VY2 VOUTN VCM +VH -VH VCM VREF -VREF VCM T1T2T1T2 Figu e 3. Timing diag am o he asynch onous sigma del a modula o wi h a cons an inpu . The du y cycle o he ADSM is gi en by he a io o ising ( SRE )- o- alling edge ( SFE ) speed. To acili a e he equa ions, a symme ical powe supply is conside ed ( VDDAs =VDDA −VCM , VCM =0V , VSSAs =VSSA −VCM , VREF =VDDAs =|VSSAs| ; |VHL −VCM|=VHH −VCM =VH ). Du ing he T1pe iod, he ou pu ol age o he i s in eg a o VY1 ises wi h speed, gi en by: SRE1=dVY1 d =I1+I2 C1 =VIN R2+VREFR1 R1R2C1 (9) and he alling edge du ing T2: SFE1=dVY1 d =I1+I2 C1 =VIN R2−VREFR1 R1R2C1 (10) The i s in eg a o ou pu ol age swing is in he ange o : ∆VY1=VY1(mean)±|VH|I2 I4 (11) whe e VHis he compa a o h eshold ol age. Senso s 2020,20, 4137 5 o 13 In o de o ind VY1(mean) , we calcula e I3(mean) . The du y cycle o he second in eg a o ou pu VY2 is he same as he i s one. F om Equa ions (9) and (10), he alue o I3(mean) is calcula ed o mee he du y cycle equi emen s. I3(mean)=VY1(mean) R3 =−VIN R2 R1R4 (12) Fo pe iod T1, we can w i e: T1=2VHR1R4C2 R1VREF −VIN R2 (13) T2=2VHR1R4C2 R1VREF +VIN R2 (14) The en i e pe iod can be exp essed as: TPER =T1+T2(15) When a ze o inpu is applied VIN = 0, he ou pu o he ADSM is a squa e wa e wi h a du y cycle o 50%. By de ining TCas he pe iod o he ou pu signal, i can be calcula ed as: TC=TPER =2T1=2T2=4VHR4C2 VREF (16) Simila o he con en ional synch onous CTDSMs, p opaga ion delay is also an issue in ADSMs. The delay o he compa a o inc eases he e ec i e alue o hys e esis and negligibly a ec s he cen e equency o he ADSM. The e o e, he impac o he compa a o delay, τ , on he cen e equency o he p oposed ADSM can be gi en by: TC=TPER =2T1=2T2=1 C =4VHR4C2 VREF +τ(17) Equa ion (17) shows ha he cen e equency C o he modula o will dec ease o a highe delay o he compa a o , which degene a es he inpu bandwid h and linea i y o he modula o [20]. 3. T ansis o Le el Realiza ion In his sec ion, he ansis o le el implemen a ion o he ADSM will be desc ibed. Figu e 4 illus a es he ci cui diag am o he p oposed ADSM. The implemen ed a chi ec u e is ully di e en ial o minimize e en-o de ha monics, as well as common-mode noise. C1 C1 R1 VCM R2 R2 VINP VOUTP VOUTN R1 VINN C2 C2 R3 VCM R4 R4 R3 Figu e 4. P oposed second-o de asynch onous del a–sigma modula o . 3.1. Ac i e-RC In eg a o In he p oposed design, he ac i e-RC in eg a o s we e used due o simplici y, high linea i y, pa asi ic insensi i i y, as well as he o e all powe consump ion. The ideal ans e unc ion o he ac i e-RC in eg a o is gi en by: In (s) = 1 sRC =ki s s(18) Senso s 2020,20, 4137 6 o 13 whe e sis he sampling equency and kiis he scaling coe icien . The pa ame e s o he esis o s and capaci o s we e designed o achie e a high cen e equency acco ding o Equa ion (16). The lowe limi o R and C is de ined by ma ching conside a ion and maximum cha ging cu en in he case o R . The uppe limi o R is se by he allowed he mal noise le el, which i sel is ixed by he o e all dynamic ange equi emen s. Finding op imal R and C was also con i med by beha io al simula ions in MATLAB/Simulink, as well as a ia ions o abou 20% in he RC ime cons an . The R and C alues we e R1 = 650 k Ω , R2 = R4 = 500 k Ω , R3 = 357 k Ω , VH = 90 mV , C1 = C2 = 2 pF, and IR = 1 µ A. I can be calcula ed om Equa ion (16) ha C = 1.39 MHz, and om Equa ion (11), VY1(mean) = (0 ± 90) mV. The alidi y o hese esul s was e i ied in MATLAB/Simulink and is shown in Figu e 5. −0.1 0 0.1 VY2 (V) −0.1 0 0.1 VY1 (V) 103104105106107 −150 −100 −50 0 F equency (Hz) Powe spec al densi y (dB) 12 12.5 13 13.5 −0.5 0 0.5 Time (µs) VOUTN (V) (a) (b) C = 1.39 MHz 0.36 0.36 Figu e 5. MATLAB model simula ion esul s: ( a ) iming diag am and ( b ) equency spec um o he ADSM o VIN = 0 V. As will be seen la e , nonideali ies such as inpu pa asi ic capaci ances o he ope a ional ampli ie and he delay o he compa a o negligibly a ec he cen e equency o he modula o . 3.2. Class AB Fully Di e en ial Ope a ional Ampli ie The in eg a o s in he ADSM we e each implemen ed using he wo-s age, Class A/AB ope a ional ampli ie opology shown in Figu e 6. This opology combines a simple di e en ial pai as he i s s age wi h a Class A/AB second s age, whe ein push-pull ope a ion is implemen ed using cu en mi o s [22]. The slew- a e is limi ed only by he i s s age. SR =I5 CC1+CG6+CG13 (19) The minimum alue o he ope a ional ampli ie slew- a e can be de e mined om he alling edge speed o VY2 ( SFE2 ) acco ding o Figu e 3. The inpu pa asi ic capaci ance o he compa a o (Ccomp) should be included in he calcula ions. Thus, |SFE2|=dVY2 d =I3+I4 C2+Ccomp =VY1R4+VREFR3 R3R4(C2+Ccomp)(20) The use o PMOS inpu ansis o s makes i possible o a oid he body e ec . The compensa ion ne wo k is comp ised o capaci o CCand esis o RM, which cancels he igh hal plane ze o. Senso s 2020,20, 4137 7 o 13 VDDA VINP VINN VOUTP VOUTN VCM IBIAS M1 M3 M6 M8M9M11 M19 M15 M17 RM CCCC RM RCM RCM M18 M16 M13 M7 M14 M10 M12 M4 M5 M2 Figu e 6. Ci cui schema ic o he wo-s age Class A/AB ope a ional ampli ie . Fo he de ec ion o he common-mode ou pu ol age, wo equal esis o s we e used (RCM =500 kΩ) . The ol age be ween he wo esis o s is sub ac ed om he desi ed common-mode ou pu ol age, VCM , and scaled by he one-s age di e en ial ampli ie ha consis s o sou ce-coupled pai M 15 –M 16 , diode-connec ed loads M 17 and M 18 , and ail cu en sou ce M 19 . The main eason o using his solu ion is ha he inpu o he common-mode sense ampli ie (ga e o M 15 ) is almos cons an . The e o e, his CMFB solu ion does no limi he ope a ional ampli ie ou pu ol age swing. Table 2sums up he simula ed pa ame e s o he ope a ional ampli ie used in he in eg a o s. Table 2. Simula ed pa ame e s o he ope a ional ampli ie (CL= 3 pF). Pa ame e Condi ion Value Hys e esis |VTH|=|VTL|90 mV Time delay CL=3pF 50 ns Slew- a e CL=3pF 42 V/µs Powe consump ion du y cycle = 50% 5 µW Inpu capaci ance 0.4 pF IBIAS 2.5 µA 3.3. Compa a o wi h Hys e esis The schema ic o a compa a o using he in e nal posi i e eedback ci cui is gi en in Figu e 7. The compa a o consis s o a di e en ial pai (M 1 –M 2 ) wi h ou pu in e e s in o de o o achie e easonable ol age swings, ou pu esis ance, and di e en ial ou pu . A second, smalle di e en ial pai , M 6 –M 7 , unbalances he inpu di e en ial pai . The inpu s o he second di e en ial pai a e ied o he ou pu signals in such a way as o in oduce posi i e eedback and, hence, hys e esis. VDDA M1 M3M4 M5 M17 M8 M6M7 M9M11 M13 M15 M10 M12 M14 M16 M2 VINP IBIAS VOUTP VOUTN VINN Figu e 7. Compa a o wi h hys e esis using an unbalanced di e en ial pai . Senso s 2020,20, 4137 8 o 13 I M 1 and M 2 a e ope a ing in s ong in e sion, he amoun o hys e esis VTH −VTL can be calcula ed using [23]: VTH −VTL =2(√ID3+ID11 −√ID3−ID11) pµCOX(W/L)1 (21) I M1and M2a e ope a ing in weak in e sion, VTH −VTL =4nUT anh−1(ID11/ID3)(22) Assume ha he ga e o M 1 ( VINP ) is ied o VDDA . Wi h he inpu o M 2 ( VINN ) much less han VDDA , M 1 is o and M 2 on, and VOUTP is a VDDA and VOUTN a VSSA , hus u ning on M 7 and u ning o M 4 and M 6 . In his s a e, no cu en lows h ough he di e en ial pai s. As he ol age a he VINP inpu dec eases owa d he h eshold poin (Equa ions (21) o (22)), some o ID5 begins o low h ough M 1 and M 3 , and simul aneously, some o he hys e esis bias cu en ID8 begins o low h ough M 4 . This con inues un il he poin whe e he cu en h ough M 1 equals he cu en ID8 . Jus beyond his poin , he compa a o swi ches i s s a e. Figu e 8shows he ol age ans e cha ac e is ic o he compa a o . The hys e esis bias cu en ID8 was 6.8 µ A, and he inpu bias cu en ID5 was 10 µ A. The ou pu high- o-low h eshold, VTL , was simula ed as − 90 mV. The ou pu low- o-high h eshold, VTH , was simula ed as +90 mV. The amoun o hys e esis was 180 mV. Simula ed pa ame e s o he compa a o ci cui a e gi en in Table 3. 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 −0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 VINP (V) VOUTP (V) VTH=VCM + 0.0932 VTL=VCM - 0.0912 +VREF -VREF Figu e 8. Simula ed DC ans e cha ac e is ic o he compa a o . Table 3. Simula ed pa ame e s o he compa a o . Pa ame e Condi ion Value Hys e esis |VTH|=|VTL|90 mV Time delay CL=3 pF 50 ns Slew- a e CL=3 pF 42 V/µs Powe consump ion du y cycle = 50% 5 µW Inpu capaci ance 0.4 pF IBIAS 2.5 µA Acco ding o pa ame e s men ioned in Tables 2and 3, he cen e equency was ecalcula ed o C=857 kHz. 4. Simula ion Resul s The ADSM was designed u ilizing he 0.18 µ m CMOS Logic o Mixed-Signal/RF, Gene al Pu pose p ocess om he Taiwan Semiconduc o Manu ac u ing Company. The ci cui was designed o VDD =1V and IBIAS = 2.5 µ A. A e comple ion o he layou design, i s pa asi ic ex ac ion was Senso s 2020,20, 4137 9 o 13 pe o med o ind he pa asi ic esis ances and capaci ances co esponding o he designed de ices and in e connec s. A e pa asi ic ex ac ion, all simula ions we e pe o med using he Spec e simula o on he Cadence pla o m. The layou o he ADSM is shown in Figu e 9. The layou size is 350 ×155 µm . R1, R2, C1 Ope a ional ampli ie 1 Ope a ional ampli ie 2 Compa a o R3, R4, C2 Figu e 9. Layou o he p oposed ADSM. The ADSM ou pu bi s eam can be eco e ed by applying an ideal low pass il e wi h a cu -o equency a he signal bandwid h. When ADSMs a e used in A/D da a con e sion, a decoding ci cui is equi ed. The simples one is he sample and hold ci cui wi h a high sampling equency. The ime domain wa e o ms o he ou pu signal VOUTN o VIN = 0 V and he co esponding equency spec um a e shown in Figu e 10. The limi cycle equency o he pos -layou model o he ADSM was equal o 848 kHz and was e y close o he calcula ed alue in Sec ion 3.3 ( C= 857 kHz). The small di e ence was caused by he pa asi ic capaci ances and esis ances ex ac ed om he layou . 1 2 3 4 5 −1 −0.5 0 0.5 1 Time (µs) VOUTN (V) 103104105106107 −200 −150 −100 −50 0 F equency (Hz) Powe spec al densi y (dB) C = 848 kHz 0.56 0.56 (a) (b) Figu e 10. Pos -layou simula ion esul s: ( a ) iming diag am and ( b ) equency spec um o he ADSM o VIN = 0 V. Figu e 11 shows he simula ed spec um o he ADSM o a sinusoidal inpu signal wi h an ampli ude o (a) 100 mV (20% modula ion dep h) and (b) 500 mV (100% modula ion dep h). The co esponding spec a we e ob ained by applying a signal a IN ≤ Bandwid h/ 3 o include a leas he second and hi d ha monic inside he band o in e es . Due o his eason, he inpu equency was se o 3.125 kHz, and hen he hi d ha monic componen was loca ed in he 10 kHz bandwid h. The achie ed SNDR was (a) 91.84 dB and (b) 78.13 dB. In he second case, he signi ican SNDR educ ion was caused by highe ha monic ones.