What would be a sensible set of filter characteristics for DAC

[philpoole]

Good Good Friday Morning to you all,

I'm designing a 3rd Order Bessel Filter for my 8xOS DAC, and its great fun!
I'm using several books, but mainly Lancaster's Active Filter Cookbook (because I have forgotten how to add), and LTSpice to simulate.
Unfortunately, I'm not sure how to get the FFT mode on to give me a sensible frequency domain transfer function - probably my input needs to be white noise or something - any ideas? Probably different software, but manually changing the input frequency and measuring the output seems to be reasonable enough.

Anyway, as I said, I'm trying a 3rd Order Bessel for best group delay. However, its not optimally flat (unlike a Butterworth) so I'm trying to make use of the luxury of 8xOS making the job of analogue filtering easier and using quite a high cut off frequency - so that the filter doesn't really start dipping until it is past the audio band. So I've looked at various cut off frequencies from 50 to 100kHz.

Now, what sort of attenuation do I need by the sampling frequency (352.8Khz)? My 50kHz Bessel attenuates a 3V input down to something like 20mV at 400kHz, but is down nearly 0.7V at 22.1kHz. 80 or 100kHz is better, only dips by about 0.4V at 22.1kHz, but likewise only down to 80mV at 400kHz.

So, my question really is, what kind of attenuation do I need around Fs? I suspect a lot of the energy that far up in the spectrum is not faithfully reproduced, but can still cause IMI if not filtered sufficiently.

BTW, this is a PMD100 digital filter chip. I suspect that's important. For instance an SAA7220 has a sneaky boost at the high end of its audio band to counter this dip in the Bessel filter.

Thinking that increasing the order would make the dip worse, I tried second order, but that actually did what I was trying to avoid. So I will investigate 4th order. I'm probably not interested in steeper cut-off, more into less dipping in the audio band.

Or, am I worrying too much?

It should be easy to knock up a generic filter circuit, and mess about at some point. To see if these limitations have any actual impact in reality.

Cheers,
Phil

[sidiy]

Hi,
What DAC are you using? A good reading is the AN of the old AD1862. They came to the conclusion that a 7th order equiripple is needed for an 8xOS.

I think that a 3rd order Bessel with -3dB@50kHz or so is good enough. FDNR is a good topology, if you want you can add more poles relatively easy without penalties on performance.

If you need the AN (~1Mb) send me an e-mail.

[martin clark]

That's a good suggestion. It's Analog Devices AN-394, a good read (i've just dug up my copy...)

[philpoole]

Yeah, I thought 50khz Fc would be suitable, and probably is.I was just a little surprised to see it dip so much at 20kHz.
I suspect I will try this and see how I get on with it. With my hearing it probably won't make much difference.
I think I've come across that doc in the past.I shall have another look.

And, its a TDA1541A DAC. I think the digital filter is probably of more interest than the DAC though.
(what with all DACs being equal )

(that was a joke!)

[martin clark]

I wouldn't worry about being fractions of a dB down; as it is, with 4x OS you get about 0.8dB roll-off, and wih 8xOS about 0.2dB at 20Khz, just from sinx/x.

(Hey , some people like non-OS, and that's 3.4dB down! Maybe even that's one reason *why* its so popular...)

[philpoole]

Hey Martin,

You know me! With my unscrupulous attention to detail, and want for everything to be perfect .
Well, alright, this might be my first time

[martin clark]

Well the CD player you thought had more treble than yours last time you popped round actually measures 0.85dB down at 20K, cf. theoretical 0.80db. I had to check,but that's close enough for me... esp after the interference it has been subjected to!

PS beers in the fridge if you get bored Happy Easter, fella.

[philpoole]

I think I shall have to go away and build something
Sounds like it'll do the job.

(It's also going to have the gain switch for HDCD - so I can listen to Beck again ).

Thanks for the reassurance guys.

[philpoole]

Well, I've built it.
It has some issues regarding decoupling. In some cases I have electrolytics perhaps a cm or so from some opamps, so I definitely need smaller decoupling caps closer to the opamps. I'll probably solder some directly underneath the socket. So at the moment its really not happy with fast opamps, which would be useful for the IV stage.
However, using some quaint JRC2114s donated by my old CD63, I'm rather pleased with the results.
So, it consists of an IV stage, then a passive 1st order filter, an opamp buffer, then a Sallen-Key 2nd order filter (incoporating an opamp) and a final opamp as a buffer.

I previously had issues with the passive stage, combined with the interconnect and preamp input impedance being hard to drive, so I've paid attention to removing that as much as possible, hence all the buffering. The opamps are socketed, so I can bypass one or both voltage followers if needed.

(BTW the RC filters are all 2k vs 1.xnF, but I had to use 2.2k - is that a major flaw? I'll change the resistances one day).

Initial impressions are great! There is much more detail. The sound stage is improved, and even things like percussion and rhythm seem better. I thought I'd do a quick test, a few excerpts from a selection of tracks, yet somehow ended up just listening to stuff in it's entirety. It was really interesting and enjoyable to listen to.

I need to tidy stuff up a bit, and especially improve decoupling, but for a first draft I'm rather impressed.

And I have deliberately left room for a relay for switching the gain for HDCD

[martin clark]

Excellent
I still have beer in the fridge if you want to drop by!

Regards