The usual approach to speeding up slow edges is to use a faster Schmitt-triggered buffer. So long as the trigger thresholds are low noise/jitter and have good short-term stability, then it should be possible to produce low-jitter fast edge-transitions. However, practice does not always work out like theory and it may turn out that a non-Schmitt-triggered buffer happens to work better for a particular case, but that would probably be more of an exception.
Perhaps some useful information here:
http://www.ti.com/lit/wp/slla364a/slla364a.pdf
At this pricepoint I just assumed they used a ams1117 3.3v and that will not give anything near a 123db SNR. Not even a LT3045 (a better LDO) will give us that good signal to noise ratio. I think it was Mark that pointed that out earlier in this tread.. But if you change out that regulator that Dac could otherwise turn out to be a little gem
simply changing ams1117 to lt3045 does not bring any measurable improvement for the board discussed in this thread. btw majority of Chinese boards use LT1963, apparently because of its fast transient response. actually, there is not even need for a fancy LDO, even 317 regs correctly implemented will be fine:
Nazar's Regulator: アナログ回路のおもちゃ箱
PSU for High end audio without electrolytic capacitors＿2: アナログ回路のおもちゃ箱
The links are not exactly what look like 317 regulators implemented 'correctly.' More like composite regulators made from 317 regulators integrated with shunt regulators. Does look interesting though. Main drawback is given as inefficiency.
Dunno, never said that any in particular sounded bad, just that maybe numbers aren’t always the determining factor.
I did start to try an lt3042 board in a digital player, but it didn’t function, never followed up to replace it since the existing supply works pretty good atm, still curious as to whether or not that would have been a significant improvement(clock, processor).
I did start to try an lt3042 board in a digital player, but it didn’t function, never followed up to replace it since the existing supply works pretty good atm, still curious as to whether or not that would have been a significant improvement(clock, processor).
yes, it consists from a series reg and a shunt where 317 stabilizes the input for the shunt and provides a reference voltage. the simulation results (and measurements) are outstanding. since integral regulators need some decorations anyway, for instance, "LDO-op amp buffer", then why not "317-3 transistor shunt"? and do we care about the efficiency here?
more on the author's site: S-Audio Systems - "Стабилизатор назара", варианты
Oven control is for frequency stability, not low phase noise. Jitter is another term for describing phase noise. Best bet are clocks specified as ultra-low jitter, and specified for low jitter or low phase noise, particularly at offset frequencies of 10Hz and even 1Hz. Low jitter or low phase noise specified up at 1kHz or even worse 10kHz is not enough to make some prediction about suitable performance for audio DAC use.
Last edited:
Sure thing, so for this particual OCXO figures at those frequencies are much better than for Cyrstek ...
Thanks. Do you think those are necessary to make this clock working? or it is just enough to connect 3.3V, GND and output?
Been listening to a 9028pro Ebay board using the LT1963. In stock form, this board blows away this 9038Q2M dac board with an LT3042 for AVCC. Now remember this an apples oranges comparison so bear that in mind. However, I now can tell immediately that the single OP amp 9038Q2M badly needs a proper IV stage. Secondly if you add in an LT3042/5 stage for AVCC without adding a proper IV stage, it is a wasted effort. So if you intend to use this cheap 9038 board, you gotta commit to making up your own IV stage as well as an OP amp PS for AVCC. In that form it might be superior to this 9028 Ebay DAC. I had version 1.04 of the 9038q2M and I blew it up while making an outboard IV stage and Op amp AVCC supply.
The stock as delivered ebay DAC using the LT1963 for AVCC using the stock 317/337/5532/*5534 sounds a lot more refined than this 9038q2m when modded with a LT3042, LME49720 op amp driven with a HP low noise low Z Sulzer power supply for the linear stages. Both boards were using filtered AC through a CORCOM filter.
When I add the LME49720/LME49710 op amps to the 9028pro DAC, it removes some of the thickness of the 5532/5534 and gains transparency and removes midrange veiling. I have yet to mod the AVCC and add the Sulzer PS to replace the 317/337. But as it is now.....it sounds sweet with the only weaker point in that the front to back is a little shallow (AVCC related based on my experience with the 9038Q2M) and I can "hear" the slightly compressed dynamic contrast which I think is the signature of the 317/337. Low end is not a problem with the 9028pro board like it was with the 9038Q2M board. I will make a new thread eventually outlining what I am doing after experimenting and learning with the now "smoked out" 9038Q2M board.
Hi Mikett, Good to hear from you. Of course you are quite right that going with a proper AVCC supply and IV output makes a really big difference in SQ. Don't forget to use good tight tolerance thin film resistors and C0G caps in the signal path. Also decouple power rails at each chip directly to a very close-by ground plane. If using leaded decoupling caps, ground them separately to keep inductance low, or maybe use a flat, wide-as-reasonable piece of copper foil as a low inductance ground strap for both. Keep it short is all.
Still not quite where it needs to be at that point, but it sure is better.
Add an ultra-low jitter clock (with fast enough edges), then add upsampling, and select the minimum group delay reconstruction filter.
Assuming proper power supplies by that point and one is probably good to go. (according to ESS, separate regulators for digital and analog power pins, and another one is suggested for any microcontroller - they don't have to all be super quality except for AVCC of course, they just need to be reasonably good, clean, separate, and hopefully close to the pins they regulate - the clock may be considered as analog, as is VCCA).
The only other possible things I can think of offhand to do would be to fix the reconstruction filter with an outboard DSP, and drop the master volume level send from Windows (or other OS) by 3.5dB to help prevent intersample overs. (The external DSP would probably be overkill, but maybe I will continue to think about it.)
Still not quite where it needs to be at that point, but it sure is better.
Add an ultra-low jitter clock (with fast enough edges), then add upsampling, and select the minimum group delay reconstruction filter.
Assuming proper power supplies by that point and one is probably good to go. (according to ESS, separate regulators for digital and analog power pins, and another one is suggested for any microcontroller - they don't have to all be super quality except for AVCC of course, they just need to be reasonably good, clean, separate, and hopefully close to the pins they regulate - the clock may be considered as analog, as is VCCA).
The only other possible things I can think of offhand to do would be to fix the reconstruction filter with an outboard DSP, and drop the master volume level send from Windows (or other OS) by 3.5dB to help prevent intersample overs. (The external DSP would probably be overkill, but maybe I will continue to think about it.)
Last edited:
There is no way of knowing as the phase noise is specced for 10MHz.
Alomst ALL of these OCXO's have specs for 10MHz because they are always
superior to the 100MHz spec.
It is very difficult to find very low phase noise, especially close in to carrier
for 100MHz oscillator. To make a valid comparison add 20dB to the 10MHz
(OC)XO phase noise and the picture will start to become clear. 10MHz is
close to optimum for XO design, especially if you want good close in to
carrier phase noise.
Terry
Paralleling various caps with an MLCC for PS bypass can result in a nice tank
circuit. You have to be very careful what you parallel.
I have measured small form factor electros that have good performance way
past 1MHz. They also get very lossy at these high frequencies which is good.
It's always worth swapping your super low Z MLCC / Tant cap arrangement
for a single high quality electro such as Panasonic FM or Elna Silmic and have
a listen to see where things sit.
The parallel cap arrangement often results in added 'detail' or HF 'air' which
may not be on the recording as opposed to the single cap which just sounds
more natural.
PS bypass on these high speed DS DAC's is tricky to say the least.
BTW - yes I know what ESS recommends...
T
Of course, that is well known. However, using manufacturer's recommended bypass is usually wise. If they say to keep leads very short, it is exactly to help avoid problem with self resonance. It they say to use a ground plane, same thing.
It you have an opamp that can oscillate at 4MHz and is sensitive to decoupling up there, probably not wise to use a cap that stops only a little above 1MHz.
Always?
Seems doubtful. Trying to tune the sound of an opamp by using caps that are less than optimal for usual bypass purposes is probably not the right way to make a circuit work if accuracy is the goal. If a pleasing sound is the goal without too much concern for accuracy, then of course, anything goes.
This discussion we are having reminds me of something that happened when I was working on my DAC mods. I had two HPAs for listening, both based on LME49720 (dual or singles) opamps and LME49600 buffers. One might think they should sound the same but they didn't.
HPA#1 used an 49720 in a single pcb and was optimized for low distortion and high speed. HPA#2 was dual mono-blocks with 49710 opamps and film bypass caps, and not a lot of them.
Of course #2 had better stereo separation, but it was also smoother and nicer sounding. #1 had, I don't know, maybe what you might call added detail and air, and didn't sound as good. That was puzzling because #1 should have been more accurate.
Turns out #1 was more accurate. I compared the DAC with both HPAs to a Benchmark DAC-3 and I could tell I still had some work to do, there was still some distortion I should be able to improve (not fully to DAC-3 level, but improve some). I was able to reduce the distortion some more, and then HPA#1 sounded better and more detailed, but in a very clean, natural way much like DAC-3. HPA#2 sounded ever so slightly muffled, lacking detail, and with slightly audible group delay (it sounded like to me anyway).
Up until that point I didn't get why what should have been more accurate, #1 didn't sound better than #2. It turned out that #1 had just been accurately telling me about my remaining distortion. Good thing I found it and fixed it. Otherwise I might have decided #2 really was better.
Actually, #2 was a better bandaid, but not what I was really looking for.
So, when I hear stories about tuning opamp sound using atypical bypass caps, I think of bandaids which is likely what is going on. Maybe not, but if using bypass caps to tune sound then I would at least not feel done until I figured out if I was just fooling myself with a bandaid or if I really found an improvement in accuracy.
OTOH, if we are talking about going for pleasing sound without regard to accuracy, or with some accuracy but also with a little touch of spice, say, then any tricks are okay with me. I will use any trick I can find to make my electric guitar sound better, or to record better sounding vocals, or to make a better sounding mix, any of that stuff.
But for a reference DAC then I want accuracy and I want to be real sure no fooling myself. That's all. DAC-3 as a point of comparison is very useful in that regard. So are measurements. Measure and listen is best whenever possible.
But, for some of these DIY projects we don't have a lot of expensive gear so we have to use the best methods we can come up with under the circumstances. Fooling one's self can be pretty hard to avoid of no measurement equipment and no reference DAC for comparison either. In that case and if interested in accuracy probably the best we can do is follow good advice of the manufacturer, measurements by people who are equipped to do them, and follow good engineering practices such as proper layout, use of ground planes where indicated, proper power pin bypassing, and so on, etc. That's how I like to approach it anyway.
Last edited:
clock buffer circuit
Is it just putting a schmitt-trigger chip into signal path between clock and DAC?
I compared Crystek with NDK - NDK has rise/fall time of 6ns and crystek CCHD 575 2ns. What would be a "low value" achievably with a Schmitt Trigger circuit?
Is it just putting a schmitt-trigger chip into signal path between clock and DAC?
I compared Crystek with NDK - NDK has rise/fall time of 6ns and crystek CCHD 575 2ns. What would be a "low value" achievably with a Schmitt Trigger circuit?