Current-feedback instrumentation amplifiers

1. Field of the Invention

The present invention relates to the field of current Feedback instrumentation amplifiers.

2. Prior Art

The notation G m as used in this disclosure refers to a transconductance amplifier, used in plurality to construct and instrumentation amplifier.

Instrumentation amplifiers (IA) are frequently used in sensor interfacing and current sensing. Current feedback instrumentation amplifiers (CFIA) gain favor because they can combine low noise and low supply power with an intrinsic high CMRR (common mode rejection ratio). The combination of low noise and low supply power is only true if their differential input stages are not being degenerated. However, non-degenerated input stages exhibit a mismatch which over temperature is not better than 2% untrimmed and 0.2% trimmed. The inaccuracy of these amplifiers directly depends on the gain mismatch of the two input stages. For most Instrumentation Amplifier applications, this inaccuracy is too high.

FIG. 1 is a block diagram of a prior art chopper current feedback instrumentation amplifier. The first input stage G m21 connects to the input through chopper Ch 21 , while the second input stage G m22 connects to a feedback attenuator R 1 and R 2 at the output through chopper Ch 22 . The ratio of the gains of the input stages divided by the attenuation of the output attenuator determines the overall gain.

The voltage-to-current functions of the input G m 's isolate the common mode input voltage from the common mode output voltage. This results in an intrinsic high common mode rejection ratio without the need of a resistor bridge that needs to be trimmed.

The input stages will have offset voltages V os21 and V os22 . When choppers Ch 1 and Ch 21 and Ch 22 are placed around the input stages, the average offset will be greatly reduced and the CMMR will be greatly enlarged (the offset V os1 of G m1 is divided down by the gain of G m21 and is of no substantial consequence). However, a ripple will result in the output voltage. A Ripple Reduction Loop (RRL) can be applied to kill the ripple. For this purpose, two capacitors C s31 and C s32 sense the ripple voltage at the Instrumentation Amplifier output. Their current is synchronously demodulated by a chopper Ch 3 , and the demodulated current is integrated on the capacitors C 41 and C 42 connected around the G m4 , or alternatively, one capacitor connected across the differential output of a current buffer CB4. The output voltage of the integrator represents the amplified offset. This offset is fed back to the output of the input stages through the correction amplifier G m3 . If the loop gain is high, the offset and ripple nearly completely vanish. The offset and ripple can both be reduced by a factor 100 to 1000.

FIG. 1 is a block diagram of a prior art chopper Current Feedback Instrumentation Amplifier (CFIA).

FIG. 2 illustrates one embodiment of the Gain-Error Reduction Loop (GERL) of the present invention in a chopper Current Feedback Instrumentation Amplifier.

FIG. 3 a illustrates a general digitally-assisted gain error reduction loop that may be used in the chopper Current Feedback Instrumentation Amplifier of FIG. 2 .

FIG. 3 b illustrates a specific implementation of a digitally-assisted gain error reduction loop that may be used in the chopper Current Feedback Instrumentation Amplifier of FIG. 2 .

FIG. 4 is a circuit diagram for the input G m21 and G m22 that may be used in the current feedback instrumentation amplifier of FIG. 2 .

FIG. 5 is a schematic illustration of an oscillator that may be used to generate the chopper drive signals for an embodiment of the present invention.

FIG. 2 illustrates one embodiment of the Gain-Error Reduction Loop (GERL) of the present invention in a current feedback instrumentation amplifier. For obtaining an optimal combination of low noise and low supply current, the input stages are composed of differential transistor pairs without resistive degeneration, and exemplary embodiment of which will subsequently be described. However, as previously mentioned, such input stages will show a substantial gain mismatch, which over temperature, may be of the order of 2% untrimmed or 0.2% when trimmed. In order to reduce the mismatch, Dynamic Element Matching (DEM) is applied to the input stages. This may be performed by 4 input choppers Ch 21 through Ch 24 which connect the input stages G m21 and G m22 to the input and via the attenuator R 1 /(R 1 +R 2 ) to the output, respectively. After each full chopper clock cycle, or after a number of full chopper clock cycle (2 being illustrated in FIG. 2 ), the connections to the input and feedback are interchanged, effectively swapping G m21 and G m22 and an equal number of additional full chopper cycles are executed, with the sequence being continually repeated with the same or different numbers of full chopper clock cycles. The straight through input connections of G m21 and G m22 are performed by engaging choppers Ch 21 and Ch 22 while choppers Ch 23 and Ch 23 are inactive (at least their outputs are open circuited, i.e., present no load on the output of the other two choppers), and the swapped connections of G m21 and G m22 is performed by engaging choppers Ch 23 and Ch 24 while choppers Ch 21 and Ch 22 are inactive Alternatively, two choppers and two multiplexers could be used for the same function. This swapping of G m21 and G m22 greatly reduces the average gain mismatch (more than a factor 100). In particular, with the swapping, typically at a frequency that shall subsequently referred to herein as f DEM , the average gain in the input path and the feedback path are nominally both the average gain of G m21 and G m22 , respectively. However, a mismatch in gain between G m21 and G m22 will result in a ripple in the output of the current feedback instrumentation amplifier at the frequency at which G m21 and G m22 are swapped, which is the DEM frequency f DEM .

It can be seen from the above that the word “swap” in its various forms is used herein and in the claims to follow in the context of changing the inputs of G m21 and G m22 from the differential input of the CFIA and the differential feedback, respectively, to the differential feedback and the differential input of the CFIA, respectively, and vice versa. In cases where the positive and negative connections of a differential output of one element need at times to be controllably connected to the negative and positive connections of the differential input of the next element in the circuit, this is referred to herein and in the claims to follow as reversing the interconnection of the sides of the differential signal, or in terms of “polarity”, and not as swapping the sides of the differential signal.

To remove the ripple resulting from the mismatch of the gains of G m21 and G m22 described above, a feedback loop is applied to continuously correct the gain mismatch of the input stages G m21 and G m22 , similar to the removal of offset by the ripple reduction loop. The capacitors C s41 and C s42 in the gain error reduction loop sense the output gain ripple caused by the dynamic element matching function. The current through the capacitors is synchronously demodulated by chopper Ch 4 at the DEM frequency f DEM and integrated on the capacitors C 61 and C 62 around G m6 . The output voltage of G m6 represents the product of the mismatch in the gain of G m21 and G m22 and the signal amplitude, and thus is also signal polarity dependent. The correction of the gain mismatch is realized by correction amplifier G m5 , which varies the tail currents of the two input stages G m21 and Gm 22 by its differential output current to equalize the gains of G m21 and G m22 . If the DEM loop gain is high, the output DEM ripple nearly vanishes, the gain mismatch automatically greatly reduces, and the overall gain is highly accurate.

As noted above, the DEM loop gain is a function of the signal amplitude, and polarity. Therefore, a comparator C 7 is used to sense the polarity of the output signal and correct the loop polarity by driving chopper Ch 5 which reverses the interconnection of the sides of the differential signal when required. If the signal is too small, the loop gain will be low because the ripple is low and the correction is weak. On the other hand, a low output f DEM ripple does not matter too much. Techniques to keep the output signal accurate at low signal amplitudes can be applied, as shall be subsequently be described.

The gain error reduction loop's gain is zero for zero input. In this state, leakage causes the gain error correction factor stored on capacitors C 61 and C 62 to drift with a time constant of several seconds, meaning that the gain error reduction loop has to re-settle whenever a finite input signal re-appears, which re-settling will occur, for example, within 10 ms for a 20 mV step input at a current feedback instrumentation amplifier with a gain of 200.

To avoid the need for re-settling, a leakage-free or drift free digital integrator can be used. The result is a general digitally-assisted gain error reduction loop shown in FIG. 3 a . In this embodiment, the analog integrator (G m6 and capacitors C 61 and C 62 of FIG. 2 ) is replaced by an ADC (analog to digital converter), a Digital Integrator and a DAC (digital to analog converter). The Digital Integrator in this embodiment receives a digital value from the analog to digital converter (ADC), which may be a single bit, or a multiple bit signal for faster settling of the loop, and is clocked by the clocking signal labeled f DEM /N.

The Digital Integrator is provided with a dead zone (output of the Digital Integrator is frozen) by an enable output of three level comparator Q 2 disabling the Digital Integrator when the output of the CFIA is in a region around zero, which avoids integrator drift for as long as the output of the CFIA remains in the region around zero. Also of course, in this embodiment, the positions of the choppers CH 9 and CH 10 may be reversed without affecting the operation of the gain error correction loop.

FIG. 3 b illustrates a specific implementation for the generalized digital version of FIG. 3 a . In this implementation, the ADC function of FIG. 3 a is done by an integrate-and-dump pre-amplifier (current buffer CB1, capacitor C 6 and switch SW 1 ) and comparator (Q 3 ). The UP/DN! Counter functions as a one bit Digital Integrator, and the 1-bit 1st order ΣΔ DAC with RC LPF low pass filters function as the DAC of FIG. 3 a . The integrate-and-dump pre-amplifier is used to integrate the demodulated DEM ripple on C 6 for, for example, seven DEM periods. The comparison is then made by comparator Q 3 . At the beginning of the next DEM period, SW 1 resets the voltage on C 6 . The comparator's output increments or decrements the 10-bit counter, whose output drives the DAC. The 3-level quantizer Q 2 controls the polarity of the loop via CH 10 and also ensures that the integrator (counter) state is “frozen” for small (in one embodiment, <8 mV) current feedback instrumentation amplifier output signals. In this context, small means that the gain in the ripple reduction loop is or is approaching too low a value to obtain meaningful input amplifier gain corrections for further ripple reduction. Thus an integrator hold mode is created between the positive gain and the negative gain conditions, i.e., in a region around a zero output of the CFIA. At steady state, when not frozen, the DAC's output will toggle between two LSBs. For example, 10-bit resolution is enough to ensure that the resulting tone is well below the current feedback instrumentation amplifier's noise level and that the DEM ripple can be reduced to the same level as the analog gain error reduction loop. Since the counter is updated at a rate of, for example, f DEM /8=1 kHz, the digitally-assisted GERL has a 1s (worst-case) start-up time. The outputs of the digital gain error reduction loop I GE+ and I GE− are the gain correction currents for G m21 and G m22 as shown in FIGS. 2 and 4 .

Thus, in the embodiment of FIG. 2 , the integration of G m6 is an analog integration, which is subject to drift in the near zero loop gain conditions, while in FIGS. 3 a and 3 b , the integration is a digital integration with a hold mode in the digital integration to avoid integrator drift in the near zero loop gain condition.

It should be noted that in FIGS. 2 , 3 a and 3 b , the inversion for negative gain done by Ch 5 ( FIG. 2 ) and Ch 10 ) ( FIGS. 3 a and 3 b ) can be done anywhere prior to the analog or digital integrator. For instance, an exclusive OR (XOR) gate between the comparator Q 3 and the counter of FIG. 3 b with one input of the gate being 1 for inversion and 0 for no inversion would accomplish the desired inversion. Putting the inversion after the integrator would cause an undesirable 2 delta jump in gain error with a relatively slow integrator recovery.

Also it should be noted that in the embodiments disclosed herein, a differential gain error correction signal is generated and used to adjust the gain of both G m21 and G m22 to make their gains equal. However adjusting the gains of both transconductance amplifiers is not necessary, as the gains may be made equal by not changing the gain of one of the transconductance amplifiers and simply adjusting the gain of the other transconductance amplifier to match that gain. Accordingly the feedback from the gain error reduction loop may be single ended to adjust the gain of one amplifier only and still achieve the desired object of this aspect of the invention.

The use of DEM (dynamic element matching) and the GERL (gain error reduction loop) ensures good gain accuracy, which means that the input transconductors can be implemented as power-efficient PMOS differential pairs, as shown in FIG. 4 . Since the input and feedback stages typically operate at different common-mode (CM) voltages, their CMRR (common mode rejection ratio) can be enhanced by cascoding the input transistors M 1 and M 2 with low-threshold devices M 3 and M 4 , respectively. During the DEM transitions, the common mode voltages of the G m stages (G m21 and G m22 of FIG. 2 ) change abruptly. As a result, the parasitic capacitances between the substrate and the n-wells of the input devices must be charged and discharged, causing large common mode current spikes in the input stages. To divert these spikes, the n-wells of the input transistors M 1 , M 2 , and their cascodes M 3 , M 4 are actively bootstrapped by class-AB buffers consisting of M 5 -M 10 . M 7 -M 9 act as level shifters to accommodate the bias voltage of the class-AB stages. The source followers M 5 and M 6 provide a low-impedance path to ground, while M 10 provides a low-impedance path to the supply. As a result, the bootstrap circuit effectively diverts the DEM spikes to the supply rails.

The advantages of a gain error reduction loop in accordance with the present invention over other techniques that improve the gain accuracy include the following. The dynamic element matching combined with gain error reduction loop has the potential to achieve high power efficiency. The reasons can be described as follows.

Firstly, since applying dynamic element matching (DEM) to the two input stages ensures good gain accuracy, this means that the input stages can be implemented as simple and power efficient PMOS differential pairs.

Secondly, chopping is a modulation technique incurring no noise folding. Therefore, it is preferred over auto-zeroing due to its superior low-frequency noise performance.

Thirdly, the noise contribution of the gain error reduction loop is negligible, since by way of example, G m5 =G m3 /400 by design. Furthermore, the gain-compensation current from the gain error reduction loop is injected at the tail currents of the input and feedback input stages G m3 and G m4 . Therefore, the noise is suppressed by the finite CMRR of the input stages, although it varies with different input signals. When the input signal of the current feedback instrumentation amplifier is close to zero, the differential currents flowing through the input pairs are almost zero, thus the CMRR is high (130 dB). When the input signal increases, the differential pair flowing with differential currents becomes unbalanced, thus the CMRR drops.

The current feedback instrumentation amplifier employs a continuous-time gain error cancellation, thus eliminating the need of trimming. Instead of one-point temperature trimming, the gain error reduction loop acts as multi-point temperature trimming, thus effectively removing the temperature coefficient of the gain.

For a stand-alone current feedback instrumentation amplifier, applying DEM combined with gain error reduction loop effectively eliminates the output signal-dependent DEM ripple. Therefore it can be interfaced with any kind of ADC, without the need to consider the synchronization between the amplifier and the ADC.

A ripple reduction loop (RRL) and the gain error reduction loop (GERL) work independent of each other at two different frequencies, a factor 4 apart in FIG. 2 . The DEM frequency may be ½, ¼, ⅛, etc., of the chop frequency, depending of the time-constant choices. When both loops are functioning, both the offset and gain mismatch of the input stages are greatly reduced. This makes the curved transfer functions of the differential transistor input stages lay precisely on top of each other, greatly reducing the effect of the inherent non-linearity of each differential pair of transistors in the two input stages. FIG. 5 is a schematic illustration of an oscillator that may be used to generate the chopper drive signals for an embodiment of the present invention. A single oscillator circuit with frequency dividers can be used, as all frequencies for an embodiment of the invention like that of FIG. 2 can be generated by dividing down a single frequency.

In the current feedback instrumentation amplifier of FIG. 2 , there is only one G m stage between the input stages G m21 and G m22 and the instrumentation amplifier output. Obviously additional stages may be added as desired, which states may themselves be chopper stabilized if desired.

The use of DEM (dynamic element matching) by periodically swapping the two input transconductance amplifiers between the signal input and the feedback input ensures good (average) gain accuracy. This may be advantageously used with or without chopping, the embodiments disclosed in detail herein being with chopping. Similarly, the GERL (gain error reduction loop) used with the dynamic element matching ensures good gain accuracy, which may also be advantageously used with or without chopping, the embodiments disclosed in detail herein also being with chopper stabilization. When using the gain error reduction loop with the dynamic element matching, the frequency f DEM is the frequency of the periodic swapping the two input transconductance amplifiers between the signal input and the feedback input. Finally, the ripple reduction loop cancels the effect of the zero offset of transconductor G m21 and G m21 and is only useable with chopper embodiments. Since chopping is only necessary for canceling the effect of the zero offset, the chopping frequency is not dependent on the frequency of the dynamic element matching (f DEM ) and of the DEM) gain error reduction loop, and can be higher or lower than f DEM , though should be different than f DEM .

Finally, note that while instrumentation amplifiers commonly have a differential input as in the embodiments disclosed herein, they may in fact be used with a single ended input simply by connection one input to a circuit ground, such as the circuit ground of the source of the single ended input signal. Also while the feedback input V fb has been referred to herein as a differential input, note that the output connection V oREF may well be a circuit ground, such as the circuit ground for the circuit to which the output of the CFIA is connected, thereby effectively making the CFIA output a single ended output, though clearly the output of the CFIA could also be a true differential output as is well known in the art.

Thus the present invention has a number of aspects, which aspects may be practiced alone or in various combinations or sub-combinations, as desired. While certain preferred embodiments of the present invention have been disclosed and described herein for purposes of illustration and not for purposes of limitation, it will be understood by those skilled in the art that various changes in form and detail may be made therein without departing from the spirit and scope of the invention as defined by the full breadth of the following claims.