Digital that sounds like analog

[opus111]

A few days ago I started up a thread on DIYA to talk about some of the progress I'd made on digital systems design with a primary focus on sound quality (rather than measurements). Its here:

http://www.diyaudio.com/forums/digital-line-level/223745-digital-but-not-numbers.html

The thread got beseiged by trolls of objectivist persuasion (people who take measurements as primary guarantees of audio quality) and then when it was subequently cleaned up by a mod the overall balance had clearly shifted in favour of the trolls in that some trolling posts were left behind but my rebuttals to those were deleted. Hence I've decided to continue the saga in the more conducive environment over here at WBF.

For those who aren't so interested in the earlier thread, the idea is to present some digital design principles, but also examples of designs which employ those principles. There will be schematics because DIY is the ultimate way to get a frugal system Nothing beats DIYing in terms of value for money.

 

[Steve williams]

Hi Richard

I started to read the thread but bowed out when I saw posts by a former member (fas42) here who always made people's blood boil with his outrageous comments and postulates

 

[opus111]

Yes Frank is an interesting one - he's come to similar conclusions as I have, though in his own way. Over on DIYA he doesn't seem to get up people's noses as much as I do. He was annoying to some at first but seems to have mellowed.

 

[opus111]

OK let's kick off with a fairly simple schematic for a filter. I've already posted this one on the older thread but didn't talk much about it and nobody commented on it at all. Its what's called an 'elliptic' filter - this form of filter has the steepest transition band of any known filter design. A transition band is the area in the jargon between the passband (the signal frequencies you want to keep) and the stopband (the stuff you don't want). Its designed to go on the output of a NOS DAC and starts cutting off around 17kHz. By 22kHz the attenuation is around -50dB. The filter I am hoping is going to improve the rather uneven HF of my NOS DAC (to be presented later) by chopping out the ultrasonic image frequencies that are inherent in NOS designs. As of writing, I've only built one channel and not listened to it so far. I'll post photos some time after both channels are completed.

 

[LL21]

OK let's kick off with a fairly simple schematic for a filter. I've already posted this one on the older thread but didn't talk much about it and nobody commented on it at all. Its what's called an 'elliptic' filter - this form of filter has the steepest transition band of any known filter design. A transition band is the area in the jargon between the passband (the signal frequencies you want to keep) and the stopband (the stuff you don't want). Its designed to go on the output of a NOS DAC and starts cutting off around 17kHz. By 22kHz the attenuation is around -50dB. The filter I am hoping is going to improve the rather uneven HF of my NOS DAC (to be presented later) by chopping out the ultrasonic image frequencies that are inherent in NOS designs. As of writing, I've only built one channel and not listened to it so far. I'll post photos some time after both channels are completed.

View attachment 6855

This thread for me is likely equivalent to the college student taking Algebra 1...and sitting in on Calculus. But what the hell...;0

So this is an analogue filter if i am not mistaken...which some designs use in their NOS dacs, yes? And it has to do with rolling off ultrasonic frequencies as you say, i think i got that part. The question i have is...if humans are 'technically' able to hear 20khz and some argue beyond, then why the rollof at 17khz? (which i think is probably my threshhold for all practical purposes anyway based on some half-baked listening/tone tests) Why not start the cutoff at 25khz? Just curious to learn.

 

[Ki Choi]

Don't mean to be too factious but... why bother spending time trying to mimic the analog sound with digital means when you can easily get the real thing?

Be content with what you have, and you'll have more time for music. I know...it is easier said than done.

Ki

 

[opus111]

Wow, nice to see you over here Lloyd With your questioning we should be able to make this more generally accessible to others who would like to savour the 'school-room' environment!

Excellent question to begin with - its 17kHz for a few reasons. Firstly because I can't hear up to anywhere near 20kHz any more. Secondly to make the filter practically possible. Its a requirement to cut off everything above 22kHz and if the transition band were just 2kHz (rather than 5kHz as we have here) the filter would be more than twice as complex. Its already a bit of a struggle winding 7 coils, per channel

Actually I'm not aware of any commercial DACs using such a filter. Your Zanden's filter is rather a different design because he didn't want the phase shift which goes together with this kind of filter. Me, I'm not so worried about phase - not until I've heard it that is. Have to build it to see for myself.

We don't start the cutoff at 25kHz because that's too late - the NOS DAC's image frequencies begin at 22.05kHz, half the sample rate.

Keep the curiosity flowing!

 

[opus111]

Don't mean to be too factious but... why bother spending time trying to mimic the analog sound with digital means when you can easily get the real thing?

Well I sold all my LPs when I emigrated to China nearly 8 years ago, so the 'real thing' is no longer an option for me. Also I am fairly sure that with digital we can get the real thing much cheaper. If I wasn't I'd not have embarked on this project

 

[Steve williams]

I too am like Lloyd on this one but simply put can you tell me what it is about your design that makes it sound analog

 

[opus111]

That's a really hard question to answer - one I've puzzled over for a fairly long time. I think with NOS DACs what makes them sound analog is they use the lowest possible sample rate and that means the lowest possible glitching. Its glitching from the DAC which 'greys out' the tonal colours and reduces the dynamics. Every time the DAC updates (at 44k1 for redbook) there's a glitch - running oversampled increases the rate of glitch production by the oversampling ratio. I can put what the digital sound lovers call 'detail' into one of my DACs just by changing the filter and I'll show how I do that too in a future post.

 

[Steve williams]

Richard

how long does it take for you to build something like this

 

[opus111]

Do you mean the DAC I'll talk about later or this filter - which I hope to retrofit to the DAC? The filter building actually about a day for one channel, but prior to that it took me ages to work out what the best inductors (coils) were to use. That would be worth almost a whole thread in itself! In my way of thinking inductors are the 'forgotten component' in electronics. I just never used them in my designs when I was employed as an engineer, didn't understand magnetics at all. Sure I used mains transformers but treated them just like a black box. So I've come up a bit of a learning curve in the past few months in working on this filter.

 

[Steve williams]

Both DAC and filter

 

[opus111]

Ah on the DAC I cheated by buying a Muse 4 * TDA1543 DAC and modifying it. So the DAC design evolved gradually - very hard to say how many hours went into it. It was spread out over more than a year but probably totally a week of time. By which I mean a normal working week since I waste so much time online too

 

[Steve williams]

are you planning on bringing these to market?

 

[opus111]

No, I plan to use them to disrupt the existing market. So I'd like to license the designs (gratis) for others to build products around.

Seeing as you have to dream big or not dream at all, how about this totally crazy notion that came to me earlier this week?

At present if you're a designer you have to pay Apple for the privilege of getting the digital data off your iPod into a dock of your design. That's a license that you have to pay them hard cash for. My own measure of success is this - when Apple pays me to let it stream its digital data into one my DACs

 

[DonH50]

What are the filter specs (passband/stopband IL, ripple, transition band, etc.)?

Why not Chebyshev so ripple is only in the stopband (assuming that is the choice you make instead of putting it in the passband)?

 

[opus111]

Here's a plot from LTSpice of the idealized filter. I haven't put in the real inductor losses yet which will mean the passband droops towards the transition band. I did look at Chebyshev type 2 but it didn't have a great slope of transition band, the elliptic was far better for similar complexity.

 

[LL21]

Wow, nice to see you over here Lloyd With your questioning we should be able to make this more generally accessible to others who would like to savour the 'school-room' environment!

Excellent question to begin with - its 17kHz for a few reasons. Firstly because I can't hear up to anywhere near 20kHz any more. Secondly to make the filter practically possible. Its a requirement to cut off everything above 22kHz and if the transition band were just 2kHz (rather than 5kHz as we have here) the filter would be more than twice as complex. Its already a bit of a struggle winding 7 coils, per channel

OK...got that one (for now...i think! ). Leave the filter at 17khz...i recall the rolloff of Zanden is only a few db above this level any way, by way of example.

That's a really hard question to answer - one I've puzzled over for a fairly long time. I think with NOS DACs what makes them sound analog is they use the lowest possible sample rate and that means the lowest possible glitching. Its glitching from the DAC which 'greys out' the tonal colours and reduces the dynamics. Every time the DAC updates (at 44k1 for redbook) there's a glitch - running oversampled increases the rate of glitch production by the oversampling ratio. I can put what the digital sound lovers call 'detail' into one of my DACs just by changing the filter and I'll show how I do that too in a future post.

As to the greying of tonal color and glitching...is the glitching a part or byproduct of the oversampling process? What is it? (if can you 'dumb it down' at all, that would be appreciated. )

Also, i note you use the term 'detail' in quotes...as if to imply (if i am interpreting your use of ' ' correctly, may not in fact be more detail...but simply more emphasis?) Am i correct in my interpretation of what you were writing? If so, is there NOT actually more detail? How DOES one get more detail out of a DAC?

 

[DonH50]

What is the passband ripple?

For others, in addition tp steep roll-off, another of the benefits of these types of filters (elliptic and Chebyshev) is their linear-phase response in the passband. That leads to constant group delay; all frequencies are delayed equally. The benefit is that signals' time relationships are maintained -- a pulse comes out looking more like a pulse and the relative timing is not changed. This is critical in pulsed systems that depend upon good time-domain performance. examples are radars, sonars, lidars (like radar but using lasers instead of radio), etc. Another example many may have seen is an ultrasound image.

The primary trades are related to ripples in the passband and/or stopband frequency response of the filters. Elliptics put ripples in both regions, Chebyshev in one (passband for Type 1, or stopband for Type 2). Passband ripples can be designed to be a small fraction of a dB (0.1 dB or less in many applications). The stopband ripples set the maximum stopband rejection, and the ripples repeat until higher-order effects roll off the filter, so in general a -50 dB stopband means it reaches -50 dB at multiple points instead of monotonically rolling off like a more common (in audio) Butterworth filter. You can clearly see the null and peak in the plot opus111 showed. By placing the null at an appropriate frequency, you can notch out something like the clock, or in this case the image that starts right at one-half the sampling frequency. The DAC tutorial threads presented a while ago show the images a DAC creates.

BTW opus111, inductors and transformers are alive and well in RF and microwave systems, although as you reach higher in frequency they may not look anything like the coils of wire you are used to in power and audio applications.