Audio DAC output Low Pass Filter.

[BrianHG]

You mean something like this:
https://www.aliexpress.com/item/PCM2704-USB-DAC-USB-Power-fiber-optic-coaxial-analog-output-Raspberry-Pi-Raspbian-RaspBMC-Windows-7/32715299458.html?spm=2114.10010108.1000014.4.7b844beeJaAfi8&traffic_analysisId=recommend_3035_null_null_null&scm=1007.13338.98539.000000000000000&pvid=c19462a3-d6f4-4e1b-90b9-8119ed0ea62d&tpp=1

I2S out version
https://www.aliexpress.com/item/PCM2706-Digital-audio-USB-Daughter-Card-Support-I2S-DAC-decoder-Italy-Amanero-same-size/32810819623.html?spm=2114.search0104.3.16.4f41c0847VyXnT&ws_ab_test=searchweb0_0,searchweb201602_3_10152_10151_10065_10344_10068_10130_10342_10547_10343_10340_10548_10341_10084_10083_10618_10139_10307_5711215_10313_10059_10534_100031_10103_10627_10626_10624_10623_10622_5711315_10621_10620_5722415,searchweb201603_12,ppcSwitch_5&algo_expid=3bead4c3-4593-4830-8a03-db48bc295106-2&algo_pvid=3bead4c3-4593-4830-8a03-db48bc295106&priceBeautifyAB=0


If you are making your own from scratch, it would actually cost you less to use a PCM2706CPJT @ 6$ (Buying 1 brand new IC), use the PCM I2S out, you choose whether to optoisolate, feeding a PCM5101A dac @ 2.70$ (buying 1 brand new IC) (106db snr), or PCM5102A @ 4.72$ (112db snr) which have built in re-clocking, oversampling with digital filtering to solve all your analog reconstruction filter issues.  Output is nothing more than a 470ohm resistor and a 2.2nf cap to GND.  2.1v output, no op-amps needed.   Even with the missing parts, and even with a cheap 4 channel 150MHz optoisolator, you still come nowhere near 50$ for this and roast the PCM2904 10.75$ (buying 1 brand new IC) burr-brown USB-DAC's SNR & no need for reconstruction filters with this $2.70 dac since digitally upsamples 48KHz to 384KHz with optional FIR or IIR filtering.

Actually, going this route makes this topic 'Audio DAC output Low Pass Filter.' obsolete, it's done digitally inside the DAC...
And if you use a separate PCB for the PCM2706, at any time in the future, you can upgrade to HighDef usb audio 2.0 since it is the same I2S interface, just use a different jumper configuration on the PCM5101A/2A dacs.

Example high definition USB audio board with PCM I2S out: https://www.aliexpress.com/item/CM6631-IIS-digital-interface-compatible-with-Italy-Amanero-USB-192K-32BIT/32810113692.html?spm=2114.10010108.1000013.10.4c3a48b5HzHmOi&traffic_analysisId=recommend_2088_5_90158_iswistore&scm=1007.13339.90158.0&pvid=89b16683-d7d6-4563-af3c-968bae4b24d4&tpp=1

[paulca]

It's not the chips that are the full problem.  The 2904 is only £8.  It's the fact I have no SMD passives.  So I need:

6 different types of resistor
5 different types of capacitor
3.5mm jacks
Micro USB sockets
adjustable linear voltage regulator
Crystal
Low Vf Diode
etc. etc.

If I swap the £9 chip for two which cost £4 or £5 it really won't help, plus it will move me into the realms where "plug and play" is threatened and I'll need to start fannying around with drivers.  It also makes the board more complex and easier to screw up.

Also, unless I'm mistaken you can't filter the "image frequencies" digitally.  Any and all DACs need analogue LPFs to remove the octave harmonics produced by the digital to analogue process.  This I got from a blog article on the TI website where this question was asked of one of their audio techs.  That and way back when I did sampling theorem at uni although those days are very foggy.

This project will remain simple and I'll maybe explore more complex I2S options after that.

[paulca]

So, back to the original question about low pass filters.

I found this:
http://sim.okawa-denshi.jp/en/CRtool.php

If I put in the values
68R and a 100nF cap.  I get a cut off frequency of 23kHz.  However the drop off is quite slow and it creates a phase shift in higher frequencies, though I don't think I'd care.

If I was to add 2 or 3 of these in series to create a 2nd or 3rd order will this sharpen the cut off above 23kHz?

Also, I gather the capacitor value has the most effect on the frequency, but I'm struggling to understand the role the resistor plays in this and how to pick it's value.  As I don't want to impede the output too much I was aiming for a low value, but that's a bit of a weak assumption.

[Bassman59]

If I put in the values
68R and a 100nF cap.  I get a cut off frequency of 23kHz.  However the drop off is quite slow and it creates a phase shift in higher frequencies, though I don't think I'd care.

All analog (IIR) filters work by creating phase shifts.

Also, I gather the capacitor value has the most effect on the frequency, but I'm struggling to understand the role the resistor plays in this and how to pick it's value.  As I don't want to impede the output too much I was aiming for a low value, but that's a bit of a weak assumption.

It's an RC filter. The cut-off frequency (the frequency at which the filter response is 3 dB down) is

f_c = 1 / (2*pi*R*C)

So it is trivially easy to calculate the filter response.

[tecman]

Interesting side note.  During the heyday of CD players, a big selling feature became oversampling.  In reality the issue was filtering the DAC output.  A filter for 44.1 KHz with little attenuation at 20 KHz is not an easy task to make.  Early CD players had 10+ pole filters made of of discrete analog R-C components.  Tolerance of components was a concern and resulted in pricey parts.  By oversampling a 160 KHz or higher DAC output filtered to 20 KHz only required a 3 pole or so simple filter.  Much lower cost to produce with greatly reduced tolerance sensitivity.  So much for the benefits of oversampling, other than the mfg's profit.

paul
 

[Buriedcode]

Lets not steer this topic into something about the pros and cons of Delta Sigma DAC,s oversampling etc..  We live in a wonderful time where we can buy DAC chips that have very high dynamic range, with high SNR and very low distortion for <$2.

To the OP.  I understand wanting to get things "right" before committing to a PCB design - no-one likes having to add extras on - so it is product to add foot prints for output filters, specifically opamp output filters.  You could always just not populate these if you add a foot print for a jumper - maybe a 0805 0-ohm resistor - to bypass it straight to the output.

Puting RC filters in series - using two caps and two resistors can indeed produce a two-pole filter with a steeper cut-off, but because they are not buffered, their gain is reduced because the second RC filter isn't driven by a zero-impedance output, its the output from another RC filter which generally has a high impedance.  A 2nd order filter using an opamp is generally more accurate, and you can easily make this a 3rd order filter with a simple RC on the output (which is driven by the opamp = low impedance output).

As with all things audio, one can get carried away trying to get things perfect, but then you'll never get a PCB made.  I'd say, find a similar design on TI's website, copy it for your board, with the facility to bypass this.

[paulca]

As with all things audio, one can get carried away trying to get things perfect, but then you'll never get a PCB made.  I'd say, find a similar design on TI's website, copy it for your board, with the facility to bypass this.

Thanks.  I did pick their DAC output filter, but of course they choose their latest and greatest expensive line driver amp which cost over a £1 each and RS only sell them in 5s.  Then of course it uses bizarre resistor values, so I have to buy another 3 resistors and one more cap.  The filter alone BOM was £10.

Still, it's not about price, it's about the fun.  Spending a total of £60 including the PCB for one sound card is nuts, but I would have components to make half a dozen or just the start of an SMD stock pile for other projects.  Besides, I have spent £60 on a night out before and only had a hangover to show for it.

As you say I can add in some 0Ohm link resistors and try it with and without the output filter to see what effect it has on perceptible quality and on the scope.

[BrianHG]

It's not the chips that are the full problem.  The 2904 is only £8.  It's the fact I have no SMD passives.  So I need:

6 different types of resistor
5 different types of capacitor
3.5mm jacks
Micro USB sockets
adjustable linear voltage regulator
Crystal
Low Vf Diode
etc. etc.

If I swap the £9 chip for two which cost £4 or £5 it really won't help, plus it will move me into the realms where "plug and play" is threatened and I'll need to start fannying around with drivers.  It also makes the board more complex and easier to screw up.

Also, unless I'm mistaken you can't filter the "image frequencies" digitally.  Any and all DACs need analogue LPFs to remove the octave harmonics produced by the digital to analogue process.  This I got from a blog article on the TI website where this question was asked of one of their audio techs.  That and way back when I did sampling theorem at uni although those days are very foggy.

This project will remain simple and I'll maybe explore more complex I2S options after that.

Dont forget the opamps & their associated filters.  That chip has a tiny 0.5vp-p output.  You've gone into an absurd high quantity of parts.  And no, none of TI's Burr-brown's USB audio chips do not need any sound drivers.

[Bassman59]

As with all things audio, one can get carried away trying to get things perfect, but then you'll never get a PCB made.  I'd say, find a similar design on TI's website, copy it for your board, with the facility to bypass this.

Thanks.  I did pick their DAC output filter, but of course they choose their latest and greatest expensive line driver amp which cost over a £1 each and RS only sell them in 5s.  Then of course it uses bizarre resistor values, so I have to buy another 3 resistors and one more cap.  The filter alone BOM was £10.

I'm completely lost. Exactly which chip are you using?

[BrianHG]

As with all things audio, one can get carried away trying to get things perfect, but then you'll never get a PCB made.  I'd say, find a similar design on TI's website, copy it for your board, with the facility to bypass this.

Thanks.  I did pick their DAC output filter, but of course they choose their latest and greatest expensive line driver amp which cost over a £1 each and RS only sell them in 5s.  Then of course it uses bizarre resistor values, so I have to buy another 3 resistors and one more cap.  The filter alone BOM was £10.

I'm completely lost. Exactly which chip are you using?

He's also combining this project with another, maybe, so we can't be sure of all the degrees of stuff involved.

[BrianHG]

Also, unless I'm mistaken you can't filter the "image frequencies" digitally.  Any and all DACs need analogue LPFs to remove the octave harmonics produced by the digital to analogue process.

The technique is called interpolation. You interpolate 48ksps signal to higher sample rate, and filter out interpolation-generated image frequencies, to get only 0~20kHz band. Then you DAC the interpolated signal, creating high frequency alias above 192ksps/2=96kHz, which is easier to filter out than 24kHz.

Well, in the case of my chosen $2.70 DAC, it's even easier since it interpolates to 384ksps/2 = 192khz.  The output filter is as shown in the data sheet, a 470 ohm resistor with a 2.2nf cap to GND giving you a smooth clean 0~20Khz band.  You wont match that internal interpolation FIR filter's 0~20Khz reconstruction with any 48Khz dac and analog filters unless you are willing to pay for precision inductors, resistors & capacitors and numerous op-amp stages in the neighborhood of 25$ in parts, PCB area and power supply requirements.

[paulca]

I'm completely lost. Exactly which chip are you using?

The TI DAC Output filter I found here:
http://www.ti.com/lit/an/sloa150a/sloa150a.pdf

The figure 1 schematic.

But I dropped it out after I figured I'd need to redo the maths for the gain level for my 2904   The maths looks a bit forbidding.  I'm sure I could plug it together, maybe with a bit of help.

It's probably overkill.  Anyone any links to a simpler design?

[BrianHG]

What sample rate will you be using?
What king of gain do you want?
What kind of output current?
Also, make sure Window's wont deviate from that sample rate.

[Bassman59]

I'm completely lost. Exactly which chip are you using?

The TI DAC Output filter I found here:
http://www.ti.com/lit/an/sloa150a/sloa150a.pdf

The figure 1 schematic.

Ahhh, OK.

The point of the DRV603 is that it can drive “line level” audio output from a circuit which has only a single 3.3 V power supply rail. And it does it without a DC blocking cap. “Line level” in the pro-audio world is defined as +4 dBu, which is 1.23 Vrms or 1.7 V_pk or 3.4 V_pk-pk. You obviously can’t swing 3.4 V from a 3.3 V power supply.

(For consumer audio, “line level” is -10 dBu, which is only 0.25 Vrms.)

The chip does its magic with an internal charge pump, which generates an internal negative voltage rail. The amplifiers do some level shifting so the output is centered on 0 V, and it doesn’t clip below 0 V because of that negative rail.

If you have a bipolar supply available, say ±12 V or ±15 V, then the DRV603 is unnecessary.

Every audio DAC data sheet I’ve ever seen has an example output low-pass filter design in the “Applications” section. Just use that circuit.

[paulca]

Every audio DAC data sheet I’ve ever seen has an example output low-pass filter design in the “Applications” section. Just use that circuit.

Apparently not this one or I wouldn't be here asking
http://www.ti.com/lit/ds/symlink/pcm2904.pdf

What sample rate will you be using?
What king of gain do you want?
What kind of output current?
Also, make sure Window's wont deviate from that sample rate.

Should be 48kHz.

The 2904 output is 0.6Vpp, which might just be good enough, but if I'm adding an amp for the filter impedence buffer I might as well double that.

Don't need much output current, it will drive a headphone amp line input.

[BrianHG]

Note when doing you calculations, you want at least 2.5vp-p  going into your headphones for cheap cell-phone/mp3 player grade audio output levels.  You would like this to be around 4vp-p for a kick-ass level, or an excellent level for high end headphone which may not have as much signal sensitivity.

[paulca]

It won't be driving headphones.  It will drive a headphone amp.  If I need extra gain I can always plug my stripboard baxandall pre-amp in.   Just need to be lifted to a sensible level.

Most of the oscilloscope traces I have seen from "domestic line output" or headphone ports have been between 500mVpp and 1Vpp.  I think my phone is about the highest I have seen, but I expect it is current limited for nanny state as the headphone output isn't all that loud.

Anyway I was watching YouTube last night and a obvious idea I have been missing popped up.  If I want to avoid the mathematics, I can get something else to do it for me.  LTSpice.  So, once I get a few things out of the way first I'll have a play with a few simulations.  I might start with the schematic for the TI line driver but put a more modest opamp in place, maybe with a rail splitter and play with the values to get something close to the cutoff I think I want.

[DimitriP]

Apparently not this one or I wouldn't be here asking

The PCM2904 is awfully similar to the PCM2900/2002/2906 which has a development board  with "everything" spelled out on the wierdly included schematic  (page 10)  :  http://www.ti.com/lit/ug/sbau195/sbau195.pdf

[paulca]

Apparently not this one or I wouldn't be here asking

The PCM2904 is awfully similar to the PCM2900/2002/2906 which has a development board  with "everything" spelled out on the wierdly included schematic  (page 10)  :  http://www.ti.com/lit/ug/sbau195/sbau195.pdf

Thanks.  I missed that one, I found ones for the 270x but not the 2906.  2904 and 2906 are identical expect the 2904 has the SPDIF pins unimplemented.

[paulca]

So I started laying the board out.  The filter/amps allow a nice ladder of Rs and Cs up either side, but it isn't going to fit in the case I was hoping for.  Won't be much bigger, but just a little.

Question.  When routing the audio signals around the board, between the DAC and the amps, does it matter if I run them beside each other or should I try and keep them as far apart as possible?

Alternatively, should I push them down to the backside and route them through the ground plane to minimise capacitance?

[paulca]

So, how bad is this for a first full attempt?

Composite:


Front side:


Back side:


VCom analogue ground net:


Should the headphone jacks not be on VCommon?  It didn't appear so from TI's schematic though.

[Buriedcode]

No.  The outputs from the opamps have DC blocking capacitors, so their output has no DC bias, so the output is referenced to GND.  Vcom is at VCC/2 because that is the midpoint of the DAC's scale, and is used to bias the opamps meaning one doesn't require a bipolar supply.

[paulca]

I keep wondering if I can actually place and air solder components that close together.  Maybe I should spread out the lines beside the op amps.  I can see it being hard to do the electrolytics as they didn't seem to like air soldering last time I tried and will need done with the iron.  I could do them first though and only some of them look tricky to get at.

[BrianHG]

Make sure your IC smd footprints are DIY assembly friendly.  They don't look it.
It may be just hard to tell from the photo...

[paulca]

Apparently not this one or I wouldn't be here asking

The PCM2904 is awfully similar to the PCM2900/2002/2906 which has a development board  with "everything" spelled out on the wierdly included schematic  (page 10)  :  http://www.ti.com/lit/ug/sbau195/sbau195.pdf

Interestingly I put this into LTSpice and the output of the amps is actually at -4dB compared to the input.  (EDIT: Sorry I actually reading the plot wrong) the whole signal in and out is -4dB).  So the amp has a gain of 1.  ... I think...

It does start to roll off at 10Khz, barely dropped before 20Khz when it's slope increases and gets linear (on the dB scale) around 30Khz and at 100Khz it's down at -26dB.

Additionally there is a small phase shift below 10Hz which obviously won't bother me, most likely from the DC blocking cap or the 0.0018uF on the input.  However there is a 40* phase swing  between 10Khz and 20Khz, surely that would be audible?  Although it does make me think of how one could use that to create a phase filter, always loved phaser pedals on distorted guitar.

Swapping R101 to be a 7k resistor gives me 0dB output, without modifying the filter cut off, so I always have that option.