LPF's Cut-off Frequency for Signal Acquisition with Multiplexer

[Politecnico]

Hi all,

I have a 32-ch Multiplexer with 32 signals (with maximum frequency of 10 Hz) connected to the inputs. By sweeping from ch-1 to ch-32, the output is read by an ADC through an amplification circuit. Could anyone tell me what the minimum cut-off frequency of the LFP at the ADC should be? According to the answer to the 1st question, how can I filter the 32 input signals at 10 Hz.

Thanks a lot, in advance.

 

[dick_freebird]

An anti-aliasing filter would be set for ADC acquisition
rate, divided by 2.

If you have known EMI / interferers then you may want
to trap those.

You don't need to filter for the purported baseband, but
for any "non-data" spurious stuff.

You want to beware simple filters because at this frequency
they probably involve very high series resistance, which
might corrupt data against the mux and other leakages.

Also consider the signal chain, location of the amplifier
and the channel commutation rate and sampling. The
settling time across commutation "address" changes,
is a factor in achievable accuracy. If your filters are
before the mux then this is not much of a concern
(although the act of switching, will put kickback noise
onto the signal source networks, and then you care
about settling) but if the filter is between mux and
ADC, settling needs to get done (to whatever accuracy)
before sampling starts.

Might want to draw yourself a state diagram and
apportion timing intervals to include these mux related
"timeouts" and then see what they indicate about the
lowest tolerable corner frequency.

 

[FvM]

Generally I don't see the purpose of a low-pass filter between mux and ADC.

The "maximum frequency" point should be clearer specified. If it means that each input signal should be reconstructed with 10 Hz bandwidth, but there are higher frequency components that must be filtered, you need individual antialiasing filter for each input. If there are no higher frequency components, no filter is required at all.

 

[CataM]

Generally I don't see the purpose of a low-pass filter between mux and ADC.

LPF between mux and ADC does not allow to pass the generated crosstalk between channels.

 

[mtwieg]

Having the filter after the MUX doesn't make sense to me. Any filter between the MUX and ADC should have a bandwidth high enough so that its settling time is less than the time spent sampling each channel.

For example if each signal has a BW=10Hz, then you must sample each at least 20sps. 50sps gives a decent margin. This means that your ADC will be sampling at 32*50sps=1600sps. So your filter should have a settling time of 600us or less. A corner frequency of at least 5kHz would be appropriate. This won't provide any antialiasing. That needs to be done with filters put before the MUX, not after.

 

[KlausST]

Hi,

. So your filter should have a settling time of 600us

I agree so far.
And I agree that there is no need for a filter at all.
The anti-aliasing filter needs to be installed in front of the MUX.

If there is a filter, then it needs to be fast enough to settle to less than one LSB before a new conversion is started.
So indeed the problem is rather how to speed the signal up (with an amplifier), than to slow it down with a filter.

A filter in this place creates cross talk errors instead avoiding it.

Klaus

 

[FvM]

LPF between mux and ADC does not allow to pass the generated crosstalk between channels.

I suggest to consider two cases:

- crosstalk inside the 10 Hz signal band. Unfortunately it can't be filtered out without distorting the signal
- crosstalk outside the signal band. The respective input signal components cause aliasing with the mux sampling process and must be filtered before the mux.

 

[Politecnico]

Thank you for everyone's advice. I really appreciate your helps.

First of all, a bit more clarification about the problem:
- The circuit is designed for the acquisition of biological signals. so, the input signal may vary from 1 Hz to 10 Hz and obviously, they are very noisy.
- Since the circuit is going to be implanted, and so, I have very strict size limitations; there is no possibility to add separate filters at the MUX's inputs.
- Assuming we have only one channel of signal acquisition and so, there is no MUX. The signals goes through the amplification stage, is filtered by a BPF at 0.5-12.5 Hz and then sampled by ADC. However, some issues appear when we have 32 channels of signal acquisition multiplexed to the amplification stage.
- ADC's output is transmitted to a back-end device for real-time monitoring / storage.

So, now the main issue is that how the BPF and what the filter's BW should be to eliminate noise, 50/60 Hz electric hum, ...

Thanks.

 

[Enrocinu]

Hi. Even i'm late for giving a view poin about the thread, i think that the ADC characteristics are important: the sampling frequency but also its resolution and dynamic range!
What is the range of acquired signals? (the ADC input SNR) to set up a suitable conditionning circuit before the ADC.
Best regards

 

[FvM]

I presume each channel will be sampled with a rate according to the Nyquist criterion, e.g. 25 - 30 samples/s. A 32 channel system will need an ADC rate of 800 - 960 Hz to sample all channels continuously.

Each mux input is acting like an ADC with 25 - 30 Hz sample rate and needs a respective anti-aliasing filter. Otherwise 50/60 Hz noise and it's harmonics will be mirrored into the signal band.

 

[Enrocinu]

However, some issues appear when we have 32 channels of signal acquisition multiplexed to the amplification stage.
So, now the main issue is that how the BPF and what the filter's BW should be to eliminate noise, 50/60 Hz electric hum, ...

I can understand that the main problem is the noise level present in the analog part before the ADC stage:-x. Improving your signal (reducing the noise) starts by the correct design (and the choosen components) of the amplification stage: NF of the amplifier+ its bandwidth= it should be a low noise amplifier in your case LNA In order to increase the output signal-to -noise ratio (SNR)
the BPF as suggested before :-D and the BW is the same for all channels and it's an advantage.
anti-aliasing eliminates the harmonics and existing frequencies outside the frequency range of interest only. And frequency sampling FS determine the BW of the final output signal i.e FS/2. This is my viewpoint:shock:

 

[CataM]

The respective input signal components cause aliasing with the mux sampling process

Each mux input is acting like an ADC

The mux gives the signal to be sampled by the ADC, what sampling process has the "mux" ?. There is 1 ADC for all 32 channels and hence all signals will ultimately go through a filter. The only point I see to the "one filter for each channel" is if every channel has very different characteristics/requirements.

You can review industry standard analog input devices circuitry e.g. the S series of analog input boards from National Instruments at page 28, figure 4.1.

 

[KlausST]

Hi,

assuming a 32 channel MUXed ADC system:
* analog input1 is sampled on 1, 33, 65, 97...
* analog input2 is sampled on 2, 34, 66, 98...
* analog input3 is sampled on 3, 35, 67, 99...
and so on.

You have to take care about nyquist on channel 1: sample 1, 33, 65, 97 ... this is ensured by an anti-aliasing filter before the MUX
You don´t have to take care about nyquist between sample 1, 2, 3, 4, 5, (but this is where a filter after the MUX operates)
After changing the MUX, there needs to be plenty of time for the analog signal to settle within 1 LSB before the conversion starts.
An anti aliasing filter can (and must) not achieve this requirement, thus you generate a crosstalk. Every channel will be influenced by the voltage of the channel sampled before.

The clean solution: analog_signal --> independent_AAF --> MUX --> (no filter) --> ADC.

The "noise" on the analog inputs has nothing to do with the MUX.
But if you expect that the noise_frequency range goes beyond nyquist_frequency, then an independent AAF is essential. Without AAF you can´t get rid of the noise on the digital side.

Klaus

 

[CataM]

I am wrong with the reference above. The reference is a simultaneous data acquisition system and hence each channel has its own filter and ADC.

However, in non-simultaneous data acquisition architecture, they usually have an active filter, see figure at page 4 of a low sampling frequency module.

 

[FvM]

According to the manual, the NI acquisition module uses individual ADCs for each channel. Apparently no mux involved. In so far the situation is quite different from the present problem.

A filter between mux and ADC is only reasonable if the inputs aren't continuously sampled. That's the situation with the NI board referenced in post #12. It has 8 differential inputs, you can select one out of eight inputs and sample it continuously.

 

[Enrocinu]

A simplified block diagram would be very helpful :???:

 

[Politecnico]

Hello Everyone,

Thanks a lot for this fruitful discussion. Here is a simplified block diagram of the circuit.

 

[Enrocinu]

Well, so each channel is processed separately. (MUXout): one channel at a time is selected according to the MCU control signal (5 bits so 2^5= 32 possible values)is a peripheral functions of the MCU . The MCU includes also the CPU and Memory. the ADC output for each channel is stored in a defined adress. :shock:

 

[Enrocinu]

There is a missing output data bus :thinker: