Hello, I suppose that one of the most important thing in a post DAC filter is to maintain flat the group delay in the 20-20Khz range (-1us in the ear band).
The second filter purpose is to kill any RF noise as much as possible (let us say in the 180-300kHz band).
The third purpose is to have a flat response in the 20-20Khz range (I would put a max -1dB at 20Khz).
Just simulating with Filter Wizard | Analog Devices or other online tools, in two minutes I have designed an 8th order filter (just two chips, four opamps), with flat group delay in the ear band and very good cut of RF noise:
-0.8dB @20KHz
-100dB @300KHz
8th order Butterworth Bessel 0.10 (4 stages, two chips)
It looks very simple (too much!), and probably I could even optimize playing more with the tool, hence, I'm asking here how it will sound such high order filter (by using good and fast stable opamps like AD826 or LT13xx) and what I'm missing in this analysis.
Or should I go for a lesser filter order?
The second filter purpose is to kill any RF noise as much as possible (let us say in the 180-300kHz band).
The third purpose is to have a flat response in the 20-20Khz range (I would put a max -1dB at 20Khz).
Just simulating with Filter Wizard | Analog Devices or other online tools, in two minutes I have designed an 8th order filter (just two chips, four opamps), with flat group delay in the ear band and very good cut of RF noise:
-0.8dB @20KHz
-100dB @300KHz
8th order Butterworth Bessel 0.10 (4 stages, two chips)
It looks very simple (too much!), and probably I could even optimize playing more with the tool, hence, I'm asking here how it will sound such high order filter (by using good and fast stable opamps like AD826 or LT13xx) and what I'm missing in this analysis.
Or should I go for a lesser filter order?
Last edited:
Yes. There are at least two things.
First you can check in your sim - cumulative noise and noise gain.
Second are absent in simple sims - PCB area and parasitic capacitance.
DAC analog must resolve next main issues:
1. Yours flat response, group delay and RF rejection.
2. As small active stages as possible.
3. Good CMRR for rejecting digital part of common mode noise.
In most cases there are easiest way to provide 3-order Bessel-approximation inverting filter in one active stage.
Check page 22:
http://www.ti.com/lit/an/sloa054e/sloa054e.pdf
R1 must be relatively low for keeping noise control, say most voltage-output DACs could drive 2-4 kOhm, while with current output DACs R1 will be omitted and R2 will set the conversion gain. With modern high grade chips having up to 16 mA output current (stereo mode of ESS Sabre) there will be another problem - high opamp load in orders of 200 Ohms. You will need empower opamp output with highly-linear buffer for keeping THD at orders of DAC output.
I want to test few old Sigma Delta like PCM1710 (in this case I think the 4th order is enough) and ....CS4303
which has differential outputs and no analog filter on board...Both chips ygg-it mentioned have eight-order digital oversampling filters on board, so to suppress images, there is no need to have suppression at 22.05 kHz, rather at 300 kHz and higher like he or she wrote in the opening post (assuming 44.1 kHz sample rate).
However, the CS4303 has a quantization noise spectrum that increases by 100 dB between 20 kHz and 90 kHz, so it is probably a good idea to use a high-order filter that cuts off at or just above 20 kHz. Cirrus recommends a sixth-order Butterworth-like filter with the first pole realized with passive circuitry to avoid slew rate problems. (A sixth-order Butterworth filter has only complex poles that you can't realize with a passive RC network, so they had to resort to a non-standard filter that behaves Butterworth-like.)
The problem with Bessel filters of any order is that the roll-off starts quite slowly. You only approach the asymptotic slope (-20 n dB/decade where n is the order) a couple of octaves above the -3.01 dB point. Butterworth filters are much better in that respect, but have worse phase characteristics. Transitional Bessel-Chebyshev filters (as derived by DeVerl S. Humpheries) are an interesting compromise, but they are not very well known.
However, the CS4303 has a quantization noise spectrum that increases by 100 dB between 20 kHz and 90 kHz, so it is probably a good idea to use a high-order filter that cuts off at or just above 20 kHz. Cirrus recommends a sixth-order Butterworth-like filter with the first pole realized with passive circuitry to avoid slew rate problems. (A sixth-order Butterworth filter has only complex poles that you can't realize with a passive RC network, so they had to resort to a non-standard filter that behaves Butterworth-like.)
The problem with Bessel filters of any order is that the roll-off starts quite slowly. You only approach the asymptotic slope (-20 n dB/decade where n is the order) a couple of octaves above the -3.01 dB point. Butterworth filters are much better in that respect, but have worse phase characteristics. Transitional Bessel-Chebyshev filters (as derived by DeVerl S. Humpheries) are an interesting compromise, but they are not very well known.
Last edited:
In the previous post transitional Bessel-Chebyshev should read transitional Gaussian-Chebyshev, also known as Gaussian-to-x dB.
All in all, if I had to design a filter for a CS4303 and if I worried more about phase than about amplitude response, I'd go for an odd-order, probably seventh-order, Gaussian-to-6 dB filter, and I would use a simple passive RC network immediately after the CS4303 for the real pole of the filter.
If you care more about amplitude than about phase response, you could either use the filter recommended by Cirrus or an odd-order Butterworth filter.
All in all, if I had to design a filter for a CS4303 and if I worried more about phase than about amplitude response, I'd go for an odd-order, probably seventh-order, Gaussian-to-6 dB filter, and I would use a simple passive RC network immediately after the CS4303 for the real pole of the filter.
If you care more about amplitude than about phase response, you could either use the filter recommended by Cirrus or an odd-order Butterworth filter.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.