Analog stage for Ozone DAC
Posted 28th November 2013 at 11:55 PM by abraxalito
Updated 12th December 2013 at 07:53 AM by abraxalito
Updated 12th December 2013 at 07:53 AM by abraxalito
A few weeks ago I built an almost completely passive analog output stage for my DAC/filter front-end and found it sounded noticeably quieter than my active ones which have been using AD605 and AD8129. The only active component in the experimental stage was a single emitter follower buffer run in classA - voltage amplification was achieved by a much higher than usual I/V resistor and passive filter specially designed for the higher impedance.
The comparison left me rather puzzled as to what was clouding the active stages. It turns out it was inadequate power supply decoupling - the fix is an abundance of ceramic capacitors soldered across the top of the chip.
On the AD605, the feedback resistor in low gain mode is a relatively low 820ohms which presents something of a challenge. Also I'm using an inverting stage following with a similar value input resistor. While the low combined values don"t present any significant challenge to the output stage itself (which is rated for 40mA) its clear that the power supply suffers. I was already using a shunt regulator with electrolytic bypass, so it seems this listening result indicates such arrangements don"t present a low enough impedance at HF and are in need of supplementing with ceramics.
I've now built two AD605 stages with such decoupling and the effect is consistent. Also I have one DAC using AD8129 which has also benefitted in a similar way. Necessary values are yet to be determined - but initially its looking like at least 100uF may be called for on the AD605 when driving around 400ohms and less for the AD8129 where I have a 5100ohm feedback resistor.
Update - I changed the supply from the shunt (low-pass filtered) to a single emitter follower driven frrom the shunt via RC low-pass. This produced a marginal improvement in soundstage clarity, but I could have been kidding myself. Then I thought about adding a classA buffer to the outputs - this would in effect halve the load impedance seen because that AD8017 stage has a lowish Zin. But thinking again I figured that as the aim is to improve the stability of the PSU, why not just add current sources to the outputs. Then the supply rails aren't so subject to the haversine currents - in effect a CCS makes the output stage classA. So now I have 5mA two-transistor CCSs and yes, amazingly the bass definition has tightened up noticeably. I attribute this to lower LF noise on the rails as the output stage draw is now more constant.
The comparison left me rather puzzled as to what was clouding the active stages. It turns out it was inadequate power supply decoupling - the fix is an abundance of ceramic capacitors soldered across the top of the chip.
On the AD605, the feedback resistor in low gain mode is a relatively low 820ohms which presents something of a challenge. Also I'm using an inverting stage following with a similar value input resistor. While the low combined values don"t present any significant challenge to the output stage itself (which is rated for 40mA) its clear that the power supply suffers. I was already using a shunt regulator with electrolytic bypass, so it seems this listening result indicates such arrangements don"t present a low enough impedance at HF and are in need of supplementing with ceramics.
I've now built two AD605 stages with such decoupling and the effect is consistent. Also I have one DAC using AD8129 which has also benefitted in a similar way. Necessary values are yet to be determined - but initially its looking like at least 100uF may be called for on the AD605 when driving around 400ohms and less for the AD8129 where I have a 5100ohm feedback resistor.
Update - I changed the supply from the shunt (low-pass filtered) to a single emitter follower driven frrom the shunt via RC low-pass. This produced a marginal improvement in soundstage clarity, but I could have been kidding myself. Then I thought about adding a classA buffer to the outputs - this would in effect halve the load impedance seen because that AD8017 stage has a lowish Zin. But thinking again I figured that as the aim is to improve the stability of the PSU, why not just add current sources to the outputs. Then the supply rails aren't so subject to the haversine currents - in effect a CCS makes the output stage classA. So now I have 5mA two-transistor CCSs and yes, amazingly the bass definition has tightened up noticeably. I attribute this to lower LF noise on the rails as the output stage draw is now more constant.
Total Comments 6
Comments
-
Yes, this sort of thing can't be repeated enough - power supplies, power supplies, power supplies. I've consistently found that the thing that one just assumes is good enough, because that's what conventional wisdom says, is precisely that which undermines the last ounce of quality achievement. Take nothing for granted, and many times one gets gets quite a surprise, in a postive sense, as a result.
Always a good philosophy to do a complete overkill in some area, just to see if it makes a difference, and then back off, bit by bit, until a good balance is struck. You've obviously done this a number of times, :), but even when one [I]knows[/I] that this 'wisdom' should be applied it's often still easy, in the heat of the battle, to think that something is 'good enough' - and then scratch one's head because the results aren't in - until the thought dawns, maybe I [I]should[/I] try varying things in this other area ...
FrankPosted 29th November 2013 at 01:02 AM by fas42
-
Thanks for the comments Frank - preaching to the choir here though - why not go over to the "CFA vs VFA' thread and see what reception you get?
Those guys over there just don"t think power supplies are so exciting as the latest loop compensation methods.
Yes to the comments about 'conventional wisdom' - that"s always the thing we have to battle. A saying of Mark Twain that I love 'its not what you don't know that kills you, its what you know for sure that just ain't so'.
I was having one of my regular browses of taobao yesterday - it was poweramps and the thing I really love there is the internal pictures (rather like 6moons but even better as audio porn goes). All the ones I looked at were pretty much the same as far as power supplies go - nobody using a switcher, everyone using a toroid. And usually the regulation two caps - just occasionally two pairs of caps.That got me thinking how different poweramp designs would be if all those guys knew what you and I know about power supplies.
For one thing there would be use of heatpipes - putting the hot devices in the middle, surrounded on all sides by caps. Thinking just like Neo: 'caps, lots of caps'.
Going back to that thread for a moment, Waly is technically adept and espouses the conventional wisdom case very well - output device linearity is what matters so parallel them up. He hasn't taken the naive path though that more is always better, he"s seen that there"s a sweet spot due to parasitics. What he so far has not taken into account though is the effect of more paralleled output devices on the PSRR. That's not contained in the conventional wisdom, at least I've never seen it. Have you?
P.S. The beachbum's on the case here, timely remark:
But IMHO, the *biggest unsimmed evil in most PA designs is the length of cable/track from the amp to the big PSU caps*. I've simmed this extensively in the last Millenium with my home brewed linear circuit analyzer .. and even more important .. tested & confirmed the findings in 'real life'.Posted 29th November 2013 at 01:28 AM by abraxalito
Updated 29th November 2013 at 02:32 AM by abraxalito -
Regarding paralleled output devices, not really. Real life feedback for some amplifier 'gurus' has made them pull back on the automatic paralleling, so there is that "knowledge" there. The other thing that intrigues me with the technique is that the devices will be constantly slightly mismatched in their dynamic behaviour around the crossover area, because of intrinsic characters [I]and[/I] thermal variations - there will be constant variations in the current balance between the devices, and current waveforms along the various paths.
It would be interesting to try and analyse to some degree how much an influence such factors could be ...Posted 29th November 2013 at 04:38 AM by fas42
-
Wondering off the top of my head whether the reason why so many DIYers are keen on lateral MOSFETs is that their Cgd is relatively low... I also note that both Thorsten and Charles Hansen are keen on them (the TO220 ones which are obsolete now) as drivers.....
Posted 29th November 2013 at 05:01 AM by abraxalito
-
Hey Frank, this article is an interesting one, a line from it made me think of you:
Despacio: The 50,000-watt sound system designed for discerning audiophiles | Ars Technica
Dewaele would go into the middle of the dance floor, with the music at 105dB, "and the quality of the sound is so high that you can still have a full-on conversation with the person next to you."
I went on to read the comments - the usual ignogance about digital vs vinyl - but the post from the medical physicist with his MRI scanner made wading through all the **** worthwhile
Posted 2nd December 2013 at 01:12 AM by abraxalito
Updated 2nd December 2013 at 01:27 AM by abraxalito -
Great find, Richard ... and on the money! I've experienced a couple of PAs set up over the years with this quality - and it tells one so much about what is possible with well done sound.
Thanks for the link - and I'll mention it in JC's BT, John should get a buzz from the referecnes ...!Posted 2nd December 2013 at 01:27 AM by fas42
