Auto-zero amplifier cuts noise in half with few components

The Analog Devices AD8553 auto-zero instrumentation amplifier has a unique architecture in which its two gain-setting resistors have no common node (Reference 1). The IC's front stage is a precision voltage-to-current converter, where gain resistor R1 sets the magnitude of the mutual conductance. The IC's back stage is a precision current-to-voltage converter that, together with the value of feedback resistor R2, determines the overall voltage gain, G = 2(R2/R1). It will be noted that the two gain-setting resistors are independent of each other and the input stage is a voltage-controlled current source, which can reduce the component count of amplifiers with extreme noise reduction requirements.

There are two ways to reduce noise by using more amplifiers. First, assume that the random noise sources in the amplifiers are independent of each other. Second, assume that the noise follows a Gaussian distribution. When averaging the outputs of conventional voltage amplifiers, the noise can be reduced to 1/√N using N amplifiers and three times the number of resistors (Reference 2). The internal structure of the AD8553 allows a nearly unlimited number of ICs to be operated in parallel using only N+1 resistors. The internal voltage/current sources connected together can also be easily operated in parallel by connecting the corresponding inputs of more ICs in parallel (Figure 1). Microvolt input voltage bias mismatches at the parallel inputs of multiple ICs are harmless here because the output resistance of the voltage/current converter is theoretically infinite.

The net result of connecting N input stages in parallel is that their output current is N(VINP-VINN)/(2R1), or N times that of a single IC. Only one of the N IC current/voltage stages is used. The feedback resistor of that stage is R2/N, where R2 is the value corresponding to the desired voltage gain AV of a single IC. Since the main noise source in the amplifier IC is its input stage, it can be assumed that the standard deviation of the random components of the output current of N parallel voltage/current converters is sNI=sI×√N, where sI is the standard deviation of the random components of the output current of one voltage/current converter. These results are different from those described in Reference 2, where the authors achieve noise reduction by averaging multiple voltages. On the other hand, the current determination part of the common output of the voltage/current converter in Figure 1 is N times the value of a single IC. The following formula is used to calculate RSNR (Relative Signal-to-Noise Ratio), which is defined as the output current to the standard deviation of the output noise: RSNRN=(N×I)/(sI×√N)=√N×RSNR1. The meaning of the formula is that the actual circuit noise is reduced by 1/√N compared to a single IC.