Noisy chopper-controlled filter

I don't know much about switched capacitor filters. We used a 7? pole LP Butterworth from Linear. It served as an anti-alaising filter on the output of a psuedo noise source. It raised the 'out of band noise' from ~ -50dB to ~ -46dB. It's never been figured out... Intermodulationm distortion was mumbled and we moved on.

George H.

This one is an RC filter with fixed capacitors and switches in series with the resistors to vary their apparent resistance.

It is possible that noise is a known problem with switching systems of this type - in which case I would hope that there is some advice around on how to minimise it.

I'm going to try changing the switching frequency and see see what effect that has. (If I can push it above the amplifiers' pass band, it might help)

he

well

he

m?

ce

on

.uk

A problem that can be difficult to address with sampled systems is that noise can get folded down to baseband. I designed one system with a demultiplexed D/A where the sampler at 8KHz (after the D/A) would mix noise from up to 1MHz all down to the 4KHz baseband and as a result the noise was equivalent to the noise from a 1MHz bandwidth rather than 4KHz.

In your circuit you need to ensure that there is no wideband noise at the input - maybe adding a low-pass filter before the first adjustable filter stage will be necessary.

kevin

The circuit diagram is at:

There is a low-pass filter which cuts off sharply at 21 Kc/s right in front of the signal chain. It comprises a Wien bridge notch, followed by an under-damped two-pole active filter which is arranged to peak up slightly at the frequencies where the response due to the Wien network is starting to fall. The Wien network notches to about -40dB at 45 Kc/s, by which time the low-pass filter is down about 12 dB and falling at 12 dB per octave.

The noise sounds like integrated white noise, i.e. the higher frequencies are absent. As I adjust the filter, I can hear a slight phasiness under the noise as the turnover point sweeps up and down.

I've now found that the worst of the noise is coming from the current-source transistor in the pulse timing section. Slugging its base to the -ve rail with a few thousand microfarads seems to kill the worst of the noise.

The bridge was referred to ground, whereas the current source is working to the -ve rail, so any noise between the rail and ground will appear as pulse timing jitter. This, in turn, will give a variation in the voltage developed across the switch-controlled grounding resistor carrying the input bias current of each op-amp (the NE5532 has a moderately high bias current).

Excellent, It's great to get things working. Do you have a bit of resistance between the power supply and big cap? Have you tried the 'popular' cap multiplier?

I screwed up the power supply regulation on a circuit and had to run a special power line back to the supply. Grrr. But I won't make the same mistake twice...(not true, but that's a different story.)

George H.

The cap is between the base of the transistor and the -ve rail, so as to remove any noise (except the transistor's own noise) from the timing current.

I'm looking into all sorts of alternatives, but the most promising one is to replace the op-amps with FET-input types, so there is no bias current for the noise to modulate. There will still be noise-behind-signal, but with the extra capacitor it is 60dB down and will be less obtrusive. I might have a problem with the distortion and noise of a chain of TL072s compared with the same chain of NE5532s, but it could prove to be the lesser of two evils.