Sallen-key filter and SNR

Stephen,

you need to analyze the noise and see what frequency range the noise is in, is it in the passband or is it in the stopband.

What are you using to measure the SNR?

Maybe you need a simple passive Rc low pass filter at the very end of the chain to get rid of the noise in the stopband that may be created by the op-amps.

Mark

Low pass filters made with capacitors have shunt paths for stop band signals to ground. If those ground points are not pristine, they also bring noise signals into the filer. What are you using as a ground reference for the filter.

Terrible opamp! 49 nV + 0.22 pA per root Hz.

What are the resistor values like?

And why 90 KHz?

John

10 pF, 90 KHz, puts your resistor values in (roughly) the 200K ballpark. Dump the opamp bias current noise into a bunch of 200K resistors, stir in roughly 20 uV of Johnson noise per resistor, and there's your noise.

Scale things down, and get a better opamp.

John

What sort of R & C values are you using? If the C values are pretty small, that could be the cause. Recall, for example, that the noise voltage (in v rms/rt Hz) for a parallel RC circuit is sqrt(k*T/C) {this assumes that the RC circuit controls the band-limiting}.

-frank

You could try

Scaling the Rs down

Scaling the Cs up

Dump R13: it just adds noise.

Sweep the response to make sure you don't have a peak.

Use a better opamp!

How many bits in the DAC? If you have enough headroom to keep all N voice signals from clipping the dac, you may be pushing each one down into the quantization noise.

John

The same ground as the rest of the PCB. The caps that go to ground on these two stages do couple straight to the non-inverting terminal. They are 10 pF, which (at least I thought) should be a high impedance to audio.

I'll try to do what I can to isolate the ground and see whether there is any improvement.

Stephen

The caps are:

Stage 1: 68, 100 pF Stage 2: 10, 100 pF Stage 3: 10, 150 pF

where the first cap goes to ground and the second is in feedback.

The resistors are 32K, 41.25K, and 40.3K for the first 3 stages (I'm using equal value resistors in each stage).

Here is a pdf of the I/V conversion stage (the DAC outputs a differential current) and the first stage of the filter:

My signal is 0.8 Vpp over an 18 kHz bandwidth, so:

49nV * sqrt(2) * sqrt(18kHz) would be the input noise voltage in peak-to-peak terms. The input signal voltage is 0.8Vpp. So: 20*log10(9.3uV/800mV) = -98.7 dB

Are there other noise contributors that aren't reflected in the input equivalent noise spec for an op amp?

Stephen

I've just been mucking around in

and was wondering what the midband gain of the SK filler is.

martin

After the first death, there is no other. (Dylan Thomas)

Well, it's cheap. ;) I'm not sure it's the weakest link though.

I posted a link to the first two stages earlier:

It's a band-limited audio signal (2-74 kHz) representing several voice channels destined to be upconverted to an RF band. I *think* that the noise power in one 18 kHz channel is much less than the signal power at

0.8 Vpp.

Stephen

1. Your opamp is really noisy. There are better choices out there. Your opamp contributes at least 15 uV of noise over the bandwidth.
2. The resistors are a bit high in value creating more noise due to thermal noise and opamp input current noise. If you need R13 for bias reasons, bypass it. It's another noise source. BTW, R13 should be 64k if your trying to minimize DC offset errors. Try use a spice simulator and see what sort of noise you get. LTspice (free from Linear Technology) is a good choice if you don't have a spice simulator. A spice simulation of your circuit with a zero noise amplifier has about

40 to 50 nV/rtHz of noise because of your resistor values.

1. You could have PCB layout problems. Digital ground noise getting in to your analog ground, digital signal passing by the analog section, insufficiently bypassed/decoupled power supplies in the analog section (i.e. running analog and digital stuff on the same power bus), and misc layout issues could all be part of your problem.

2. You should rig up a noise test and try out different opamps and component values. A superstrip bread board will work fine.

I could have been more specific. I am trying to get you to think of what else is sharing the ground, and how it is routed across the board, and how ground currents will produce local voltage drops that show up in unintended places.

Is the audio ground shared with digital chips, etc.?

I suggest you plug in a MCP6044 higher performance opamp (especially with a 5 volt supply) and with lower noise.

A 10 pF 1% cap won't be anything like that once you put it on a board and connect it to opamps. An opamp input runs several pF, typically. The PCB pads and traces alone will add a good hunk of a pF, and FR-4 makes just about the worst capacitors on the planet.

There are filter configs that have lower component sensitivities than Sallen-Key.

John

Well... you don't state your operating level making it difficult to make any use of S/N numbers.

40nV/rtHz isn't especially quiet for an op-amp - actually it's noisier than any I can think of offhand..

Each filter stage will also add thermal noise too. Approximately what are the component values for each stage ?

Graham

I'll look into scaling the design.

The reason I started with 10 pF and 100 pF capacitors is that those parts were available in 1% values. I don't think that 1% 1000pF caps are avaible (surface mount, 0402 or 0603, but I might be mistaken there).

Trying to design a low ripple Chebyshev filter (several tenths of a dB at most) with 5% caps is challenging.

I'll first try a lower noise op amp (maybe on semi's MC33079). Thanks so much to eveyone for their suggestions.

Stephen

Keep in mind that stray trace and pad capacitances are probably way more than 1% of your 10pF capacitors.

As a 'rule of thumb' I'd avoid using R values > 10k where you want to avoid adding much noise in such a filter and that's with a higher working reference level for audio of around a Volt.

You might consider simply scaling those values by a factor of 10 for example.

Thanks - that gives me something to work with.

Well... that's 283 mV rms.

There seems to be a term of sqrt(2) in your calcualtion that gives you *peak* voltage but not peak to peak voltage.

I make the rms input noise of the op-amp over that bandwidth 6.57 uV

Which gives me -92.7 dB ! You made the error of comparing peak and peak to peak voltages. Hence the 6 dB difference. For simplicity I always normalise to rms - it's less trouble !

Well... there's input bias current noise but I doubt that's the case here.

However you're only getting 67dB at the input to the filter stages. How much self noise is there from the DAC I'm wondering ? You could simply load it with Rf and measure directly.

Graham

Seconded. I've recently become aware that Microchip do some curious stuff I wasn't previously aware of.

For example I just discovered they do gate drivers for power fets.

Graham