Correctly adjust the filter components to improve noise reduction

In DC to low frequency sensor signal conditioning applications, the common mode rejection ratio (CMRR) of the instrumentation amplifier alone is not enough to provide robust noise rejection in harsh industrial use environments. To prevent the propagation of unwanted noise signals, it is critical to properly match and adjust the components in the low-pass filter at the input of the instrumentation amplifier. Ultimately, the internal electromagnetic interference/radio frequency interference (EMI/RFI) filtering and CMRR work together to reduce other noise and achieve an acceptable signal-to-noise ratio (SNR).

For example, consider the low-pass filter implementation shown in Figure 1. The resistor sensor is connected differentially to a high impedance instrumentation amplifier through a low-pass filter network consisting of RSX and CCM. Ideally, if the CCM of each input leg is perfectly matched, the amount of noise shared by both inputs will be reduced accordingly before reaching the INA input.

Figure 1 Common-mode input filtering

When the common-mode filter capacitors (Ccm) are perfectly matched, the noise is almost completely eliminated. Figure 2 shows this result from a TINA SPICE simulation, which injects a 100 mVpp, 100 kHz common-mode error signal into the INA333 input.

Figure 2: Example simulation of complete input RC matching of INA333 common-mode filtering

The problem with this approach is that off-the-shelf capacitors have a typical tolerance of 5% to 10%, which means that if the CCM of each leg is not matched in the opposite direction, the total differential tolerance can be as high as 20%. Figure 3 better illustrates this capacitor mismatch and also shows the common-mode noise input (eN) at the output of the resistor sensor.

Figure 3 RC mismatch and common mode noise injection Common mode filtering

This input mismatch (∆C) creates a cutoff frequency error, allowing common-mode noise, eN, to differentially enter the INA input and then be gained out as an error voltage. Equation 1-3 shows the amount of common-mode noise that reaches the input:

Equation 1

Equation 2 2 Equation 3

Assuming that the frequency of the sensor signal Vsensor is much lower than the noise cutoff frequency of all common-mode filters (i.e., fC ≥ 100*fsensor), and RS1 = RS2, the magnitude of the common-mode noise signal (eN) converted to the differential noise signal (eIN) and becoming part of VIN is:

Equation 4

Equation 4 further shows that by injecting a 100 mVpp, 100 kHz common-mode error signal into the INA333 with a 10% RC mismatch at the 1.6 kHz filter cutoff frequency, the resulting error is:

Figure 4 Simulation of INA333 output error caused by common-mode filter RC mismatch (gain is 101)

Figure 5 shows a better and more common input filtering approach, which is to add a differential capacitor, Cdiff, between the in-amp inputs.

Figure 5: Adding differential capacitance (Cdiff) improves common-mode noise suppression

Adding this capacitor does not completely solve the problem, because Cdiff must be adjusted according to the following two criteria:

1. The differential cutoff frequency must be high enough to be far away from the signal bandwidth to achieve sufficient filtering stability.

2. The differential cutoff frequency must be low enough to reduce the common-mode noise to an acceptable level, so that the instrumentation amplifier CMRR can achieve residual noise suppression and ultimately achieve an acceptable SNR. Equation 5 gives the general principle for making this adjustment:

Equation 5

Figure 6 shows the VinP and VinN graphs versus two frequencies without Cdiff and with Cdiff = F. Note that without the differential capacitor, there is a difference in the magnitude of the INA333 outputs. This difference is amplified to the output as noise which ultimately reduces the SNR. When Cdiff = F, the difference between VinP and VinN is minimal.

Figure 6. VinP and VinN curves for Cdiff = 0 and Cdiff = 1 F.

Figure 7 shows the overall noise improvement at the INA333 output when Cdiff = F.

Figure 7 INA333 simulation of improved noise filtering using Cdiff

In summary, the low-pass filter placed in front of the instrumentation amplifier should have a differential capacitor that is at least 10 times larger than the common-mode capacitor. This greatly improves the filter's efficiency by reducing the effect of Ccm mismatch and turning common-mode noise into differential noise.

Next time, we will explain some of the difficulties of I2S clocks in master/slave systems to you, so stay tuned.