Discrete and Continous Discrete – sampling is required in time domain. Should follow Nyquist criteria i.e. Clock rate >> 2* BWsignal; to reduce the requirements of an anti-aliasing filter. As a result, switched-capacitor filters are limited in their ability to process high-frequency signals. Continous Time filter Advantage Signals remain continuous in time and have analog signal levels. No sampling increased speed Disadvantage: need tuning circuitry poor linearity and noise performance. Continous Time filter Integrators, summers, and gain stages are the fundamental building blocks of analog filters. Realize any rational function TF Any rational function with real valued coefficient may be factored into first- and second-order terms First-Order Filters First-Order Filters • • • • • the number of integrators = order of the desired TF real pole at a freq = 0. The pole to lie in the LHS of the s-plane, the value of 0 must be positive. DC gain, H(0) = ko/ 0 Gain at very high frequencies = H()= K1. Second-Order Filters / biquadratic filters / biquads INTRODUCTION TO Gm-C FILTERS – Integrators & Summers s V0 = integration of differential input voltage multiplied by integrator unity gain frequency 0. Fully differential Integrator IC applications signals are kept differential Better noise immunity and distortion properties. Single Vs Two capacitor Diff. Ckt Single Capacitor – Parasitic Effect First Order Filter First Order Filter First Order Filter 0<k<1 Cant use in high frequency gain ckt Fully differential first order Gm-C filter TRANSCONDUCTORS USING FIXED RESISTORS Objective : linearize the relationship between differential input voltage and output current of a differential pair of transistors input voltage,Vi, appears across the two Rs/2 resistors and Rs If VGS=0 If Vi=0 Bias Current I1 exist through Rs/2 Vi will be closer to ground- undesired Extend of relation stands to the value upto which gm can equal 1/Rs, ensuring linear realtion b/w io and Vi even with varying Vbe For linearity , bias current should be large and input signal should be small (to keep gm small) Alternative - Use opamp Using current mirrors – constant current through Q1, Q2 Transconductance using triode transistor Transconductance using triode transistor For n channel transistor in triode region, Transconductors Using a Fixed-Bias Triode Transistor VDS=0 ; ID VDS Transconductors Using a Fixed-Bias Triode Transistor Transconductors Using a Fixed-Bias Triode Transistor Constsnat current flow through transistors Q1 and Q2 so that VGS is constant Gm = 1/ rds of Q9 Transconductors Using a Variable-Bias Triode Transistor To ensure Q3 and Q4 lie in triode region V1=V2 ie Vx = Vy = Vi-VGS1 = V1- (Veff1+Vtn) ie for Q3, Q4 , VGS of Q3 & Q4 = VGS of Q1 & Q2 ensuring Q3 , Q4 is in triode region VDS = 0 To find the Gm ID in saturation Gm is not linear wrt I1 Gm can be tuned by changing I1 Transconductors Using Constant Drain-Source Voltages Transconductors Using Constant DrainSource Voltages Gm depends on constant VDS only Bias current I1 also depends on constant VDS if small VDS CMOS TRANSCONDUCTORS USING ACTIVE TRANSISTORS worse linearity performance - moderate (around 30 to 50 dB) –based on square law for ID have improved speed performance Methods Constant Sum of Gate-Source Voltages Bias offset Cross coupled differential Pair I. Constant Sum of Gate-Source Voltages To maintain Constant Gate Source Voltage Source-Connected Differential Pair Inverter Based Differential-Pair with Floating Voltage Sources Source-Connected Differential Pair Source-Connected Differential Pair Input signal varies symmetrically around a common mode voltage Linearity limited due to square-law model being inaccurate Even-order harmonics occur if the difference between two drain currents not exact. Limited to less than 50 dB linearity Adjust Gm by varying VCM In a short channel process, velocity saturation limits transconductance variation CMOS Pair – Two transistor Circuit Two transistors in active region Eqvt to a single transistor Vt-eq – Eqvt. Threshold voltage Keq - Eqvt. parameter Inverter Based ; Io=??? If Gm = 4Keq(VC1 – Vt-eq) Transconductance value can be varied by changing the control voltages, VC1 and VC2 Matching should occur only for transistors of the same type. ie, the n-channel transistors should match and the p-channel transistors should match. But there is no need to match n- and p-channel transistors to each other. Differential-Pair with Floating Voltage Sources Differential-Pair with Floating Voltage Sources VGS1 – (Vx + Vtn)+ VGS2 –(Vx + Vtn) = 0 VGS1+ VGS2 = 2(Vx + Vtn) ie. circuit maintains a constant sum of gate-source voltages even if the applied differential signal is not balanced. V1 – VGS1 + Vx + Vtn = V2 V2 – VGS2 + Vx + Vtn = V1 VGS1 - VGS2 = 2(V1 – V2) ie. the difference between the input voltages =1/2 ( the difference between the gate-source voltages) Differential-Pair with Floating Voltage Sources We know, (iD1 – iD2) = K (VGS1+ VGS2 – 2Vtn)(VGS1 - VGS2 ) (iD1 – iD2) = 4KVx(V1 – V2) ie, Gm = 4KVx II. Bias-Offset Cross-Coupled Differential Pairs Bias-Offset Cross-Coupled Differential Pairs Bias-Offset Cross-Coupled Differential Pairs O/p differential current is linear wrt the differential i/p voltage Gm VB. Bias currents through Q5,Q6,Q7 and Q8 are all square-law related to VB So, 1) Gm of Q5,Q6,Q7 and Q8 square root of the changing bias current 2) Bias current, Iss , doesn’t affect Gm but determine the maximum (or minimum) output current available. BIPOLAR TRANSCONDUCTORS 1.Fixed Transconductor with Gain Cell Fixed Transconductor – a differential pair linearized through resistor degeneration. Gain cell - allows scaling of the output current by varying the ratio of two current sources. • Larger input range, lower transconductance • Create a transconductor with a fixed value using a resistor • Tune the transconductance using a gain cell 2. Multiple Differential Pairs • Smaller input range, higher transconductance • Use 2 diff pairs to extend linear input range • Tuning done by adjusting bias current • More linearity is offered Gain Cell Transconductors - Gain Cell To execute variable transconductance, gain cell is used In a gain cell, the output current is equal to a scaled version of the input current The scaling factor is determined by the ratio of two dc bias currents. Also, the circuit is highly linear. Gain Cell Transconductors io1 = Vi / RS io = (I2/I1) io1 Gm Gain cell below the input differential pair, Gain cell below the input differential pair, 2re vs 4re in series with the degeneration resistor, RE, which might affect the distortion performance. The gain cell shown has less distortion due to finite effects and has therefore been referred to as a betaimmune type-A cell Transconductors Using Multiple Differential Pairs – Differential Pair Gm I1 Limited input range when used as a linear transconductor. when Vi > 32mVpp, the total harmonic distortion of the output current > 1% use parallel differential pairs to increase the linear input range currents of the right side transistors in each transistor pair. If the dc offset voltages,V1 , are chosen carefully, then the two current curves will partially linearize each other when they are added together, In this case,V1 is chosen to be equal to 1.317 VT to maximize the linear input range Thus the input range can now be approximately three times that of the original differential pair or, equivalently, 96mVpp and achieve the same distortion performance. To simplify the circuit Eliminate d source by approx. sizing the differential pair of transistors Let Vi=0, V1= 1.317VT, Ic of Q2 and Q4 If a two-transistor differential pair has unequal-sized transistors such that the same two currents occur for Vi=0 , then this unequal-sized differential pair will behave the same as that of an equal-sized differential pair with a dc offset applied. Size of unequal transistors k= ratio of area of Base – emitter of Q1 to Q2. If Vi=0, Q1 3.73 times larger than Q2, and similarily Q4 should be 3.73 times larger than Q3. Such transistor sizing achieves the same result as the dc offset voltages. The ratio of the two currents be 0.7887/ .2113 when their VBE are equal resulting in the area ratio. Area ratio is set to 4 for practical reasons. The final linearized transconductor For this circuit with Vi=0 , the currents through Q1 and Q2 are and 0.8 I1 , 0.2I1 resp. Active RC filter & MOSFET C Filter To realize Analog integrated filters Principle (Miller integration) : integration of current is performed upon capacitors connected in feedback around a high-gain amplifier. Miller compensation capacitor in a two-stage opamp Gm-C filters, use grounded capacitors to integrate current. Miller integration: - improves linearity - need high gain-bandwidth product in the amplifier making active RC and MOSFET-C filters slower than Gm-C filters. Opamps are capable of driving resistive loads are required in active RC and MOSFET-C filters reducing speed vs the capacitive loads in Gm-C filters. But used in BiCMOS technologies where hightransconductance opamps are available. Active RC filter Vx - the two input voltages of the opamp equal due to –ve feedback If components during +ve and –ve half cycles are equal, Vdiff = positive integration of the two input differential signals. Negative integration?? MOSFET-C filters Similar to fully differential active-RC filters, but resistors are replaced with equivalent MOS transistors in triode. Variations of MOSFET-C filters — Two-transistor integrators Four-transistor integrators R-MOSFET-C filters. MOSFET Two-Transistor Integrators Vx determine the small-signal resistance of the transistors. Vc – for tuning Vp1, Vp2, Vn1, Vn2 are balanced around VCM, Making o/p also balanced Vpo, Vno, is linear. Since the circuit is fully differential, even order distortion products cancel and two-transistor MOSFET-C integrators realize filters with around a 50 dB linearity. Four Transistor Integrator - A single-input fourtransistor MOSFET-C integrator. Improve linearity The small-signal analysis of this four-transistor integrator treat one-input integrator as a two-input integrator in which the two input signals are Vpi - Vni and the inverted signal is Vni - Vpi Effective resistance is determined by control voltage 4 transistor ckt can cancel both even and odd distortion products if the nonlinear terms are not dependent on the control signals In long channel transistors, non linear distortion terms in IDS is independent of control voltages Linear term depends on controlling gate voltage, But, all distortion terms of the ID remain equal in the pair Qp1, Qp2 & Qn1, Qn2 as VDS is equal in each pair. Due to the cross-coupling connection, both the even and odd distortion terms cancel. Not available in short-channel devices. Also, transistor mismatch limits the achievable distortion performance. ie a 10-dB linearity improvement using this technique over the twotransistor MOSFET-C integrator. R-MOSFET-C Filters - Use additional linear resistors Rq1,Rq2 - small-signal drain-source resistances of Q1, Q2 DC gain of this first-order filter is not adjustable, but can be set precisely since it is determined by the ratio of two resistors. Time constant = R2C1 can be changed by changing the values of Rq1,Rq2 by varying the control voltages. Low freq – R2 set Vx to VG – increase linearity (-90db) High freq – dec in linearity to -70db TUNING CIRCUITRY CTF - requirem of additional tuning circuitry. Why - large time-constant fluctuations due to process variations. Eg: IC, R, Gm Time constant RC or Gm-C variation??? - temperature variations Abs. component value vs relative component value Abs. vs relative value of transcondcutoance Let (i) Gmi = Kmi x Gm1. (ii) CX + CB = kXBCA Kmi – appropriate transistor, resistor, or bias-current scaling. KXB - ratio of capacitance values. KX = CX / CA Gm1 / CA… and extends to all transconductors indirect frequency tuning To tune a continuous-time integrated filter, an extra transconductor that is tuned is used and the resulting tuning signal to control the filter Transconductors Such an approach is commonly referred to as indirect frequency tuning. This tuning is indirect as matching between the filter transconductors and the extra transconductor is focussed . ie the filter is not directly tuned by looking at its output signal. Constant Transconductance No tuning – 30%tolearnce in abs. value for Gm/C. Tune Gm only if ?? Else… How – Employ ext. resistors Use a known value of external resistance and measure the step response of the filter determine the C value from step-response result Proper resistance is calculated. Use f/b ckt to set Gm = 1/ ext. resistance Frequency Tuning – Approach 1 Frequency Tuning Req = 1/(fclkCM) Gm/C A = (fclkCM)/ C A Precise frequency tuning can be achieved without any external components. Problem-1 Requires large transconductance ratios leading to poorer matching if high-frequency filters are desired. Solution - reducing fclk by increasing CM. But .. Set a smaller transconductance value for tuning circuit – need proportiante increase in fclk in filter transconductor Leads to poor matching between the filter and the tuning circuitry, resulting in a less accurate frequency setting for the filter Approach 2 – Using two scaled current sources Req = - 1/(fclkCM) diode-connected transconductor is equivalent to a resistor =1/Gm. When the average current into the integrator is zero, Gm = N fclkCM . ie the clock frequency of this circuit is N times lower than previous circuit Approach 3 – Using PLL Two continuous-time integrators are placed in a loop to realize VCO that is placed into PLL. Once ON, the negative feedback of the PLL causes the VCO frequency and phase to lock to the external reference clock. Once the VCO output is locked to an external reference signal, the Gm /C ratio of the VCO is set to a desired value, and the control voltage,Vcntl, can be used to tune the integrated filter. Q Factor tuning Q-factors of the poles of the integrated filter are tuned due to high-speed or highly-selective filtering, non ideal integrator effects and parasitic components By tuning a single time constant, can set all coefficients of an integrated filter with around 1% accuracy But if Q>1, even small errors in the Q-factor can result in large errors in the filter frequency and step response. Hence, in some applications additional tuning is needed to precisely set the Q-factors. Method -1 Tune the phase of the filter’s integrators to a 90-degree phase lag near the filter’s passband edge. Done by using tunable resistor in series with integrating capacitor The control voltage for this tunable resistance,VQ, is generated by a Q-factor tuning circuit and passed to all the filter integrators. If integrated filter is second order, the pole Q can be adjusted by changing the transconductance of the damping transconductor in the filter. Tuning Methods Based on Adaptive Filtering used in DSP applications such as model matching, channel equalization, and noise (or echo) cancellation. Adaptive-filter tuning method intended for highspeed data transmission over a twisted-wire pair pulse-shaping filter - to ensure that not too much highfrequency power is radiated from the twisted-wire channel. frequency and Q-factor loops rely on the fact that the input to the filter is a series of steps ie tune to step response of the filter Frequency tuning is done by ensuring that the zero crossing of the filter’s low-pass output occurs at the correct time period after a data transition. Q-factor tuning is performed by comparing the peak in the filter’s bandpass output with a known voltage. The peak in the bandpass output occurs at approximately the same time as the zero crossing, so both detectors can be realized through the use of two clocked comparators, which are triggered at a set time after a data transition. Frequency tuning is done by ensuring that the zero crossing of the filter’s low-pass output occurs at the correct time period after a data transition. Q-factor tuning is performed by comparing the peak in the filter’s bandpass output with a known voltage. The peak in the bandpass output occurs at approximately the same time as the zero crossing, so both detectors can be realized through the use of two clocked comparators, which are triggered at a set time after a data transition.
