Are you talking noise density or RMS noise?
If using noise density, I remember some AN measuring noise floor around 1-2 nV/SQRT(Hz) and the TPS measures 14 nV/SQRT(Hz)
the new DSD korg dac use CS4398, that like iron a lot
the same dac in the MR1000 and MR2000 DxD recorder
Pure Super Audio: KORG to release an ultra-high resolution 5.6MHz DSD DAC with affordable price tag
http://www.head-fi.org/t/632686/korg-ds-dac-10-usb-1-bit-dsd-dac
the same dac in the MR1000 and MR2000 DxD recorder
Pure Super Audio: KORG to release an ultra-high resolution 5.6MHz DSD DAC with affordable price tag
http://www.head-fi.org/t/632686/korg-ds-dac-10-usb-1-bit-dsd-dac
Last edited:
does it matter? in both ways the Sala shunt, jung shunt and for example Ackos flea variant turn out to be significantly less than 1µV RMS, meaning they are measured in nVRMS and the impedance in both cases are yet another order of magnitude different. but yes RMS, I havent bothered to work out the v/ sqrt Hz for the tps for your pass band/s (which I guess you pulled from Ian;s numbers. its a good reg, its still handily beaten in all areas of performance except price by these others
Last edited:
Well, 99% sure I wont "hear" those numbers, but if a regulator measures 1 uV RMS 10-100K, then the noise density is 3 nV which is near noise floor. Do you have links to noise density charts for the shunts?
The pass band is from the manufacturers. Everyone reports RMS numbers typically 10-100K. The noise density is approximated as a square box according to a tutorial from Analog Devices (The RMS is the square box area of the density)
Last edited:
The place you'll really hear low noise is the psu for the oscillator, but in fact, dropping noise levels everywhere is the best way to get more definition and stereo imaging.
If the load is pretty constant, like most digital lines are, then you can use an emitter-follower to drop noise down to the noise floor of the transistor, so then it becomes about how well you can make it and your choice of transistor (low noise, high gain bandwidth, low input capacitance). The only downside for this is the lack of load regulation and the output impedance is quite high ~1 to 2R. But these downsides don't matter for digital-constant-load.
I've done this many times and heard the difference. It's easy, simple, and cheap and has real sonic benefits.
If you saw what i did with the humble LT1084, which I learned about from others work on the 317, you can see I have two RC filters - this very effectively blocks high freq noise, which the regs can't do well, and the lt1084 takes care of the low freq ripple. Killing the 1~2Hz noise is still a challenge and that can mess up a clock too. I have quite a lot of source impedance because of the RC filters, but this is not significant for the application = pre-reg for digital
The ADP7104 is available in 9V @ 15uV noise. You can use some rc filters on its output to feed the base of the emitter-follower and create a nV noise 3.3V supply for digital-constant-load.
But the above is no good for analogue - that needs very good load regulation and low output impedance.
That link to Marce says, "For clocks I would recomend some very small COG caps(pFs) with 0.1uF X7R's as intermediate reservoir caps, then some 10uF X7R then a few larger elecs or tants."
Funny, I seem to remember making something like this and getting flamed for it.
I think I have to quit Diyaudio. There's a very vocal someone here who likes to be right too much, even when it is destructive and perverse.
If the load is pretty constant, like most digital lines are, then you can use an emitter-follower to drop noise down to the noise floor of the transistor, so then it becomes about how well you can make it and your choice of transistor (low noise, high gain bandwidth, low input capacitance). The only downside for this is the lack of load regulation and the output impedance is quite high ~1 to 2R. But these downsides don't matter for digital-constant-load.
I've done this many times and heard the difference. It's easy, simple, and cheap and has real sonic benefits.
If you saw what i did with the humble LT1084, which I learned about from others work on the 317, you can see I have two RC filters - this very effectively blocks high freq noise, which the regs can't do well, and the lt1084 takes care of the low freq ripple. Killing the 1~2Hz noise is still a challenge and that can mess up a clock too. I have quite a lot of source impedance because of the RC filters, but this is not significant for the application = pre-reg for digital
The ADP7104 is available in 9V @ 15uV noise. You can use some rc filters on its output to feed the base of the emitter-follower and create a nV noise 3.3V supply for digital-constant-load.
But the above is no good for analogue - that needs very good load regulation and low output impedance.
That link to Marce says, "For clocks I would recomend some very small COG caps(pFs) with 0.1uF X7R's as intermediate reservoir caps, then some 10uF X7R then a few larger elecs or tants."
Funny, I seem to remember making something like this and getting flamed for it.
I think I have to quit Diyaudio. There's a very vocal someone here who likes to be right too much, even when it is destructive and perverse.
Last edited:
The AKM looks interesting: share PCM pins with DSD, S/W volume control, Dual mono for 60 bucks, Hmmm
Indeed. I've been interested in it for a while, and now there is a low-price diy board....
How easy do you think it is to write code to control volume etc ? It is 32 bit so the volume control should be decent I hope.
Last edited:
KK, funny, marce models and measures everything he does, the software hes using will pretty much model it all as he makes the PCB design. there is nothing wrong with the parallel thing if its done correctly, you obviously started to react before actually reading what was written.
the problem is and he mentions this too, that you can set up resonances with the trace and part inductance, which can cause ringing and randomly choosing (same as just using the same values for everything) values to use on a completely un-characterized PCB is a good way to get them. I also dont see you using 0402 parts and there is nowhere on the PCB close enough for a useful low value bypass... your process seemed to be just choosing a mixture of caps and putting them where they will fit, with no measurement; can you explain the similarity?
that is what I critiqued, not the use of the parallel parts. the technique is good, but 'risky' to do blind.
the problem is and he mentions this too, that you can set up resonances with the trace and part inductance, which can cause ringing and randomly choosing (same as just using the same values for everything) values to use on a completely un-characterized PCB is a good way to get them. I also dont see you using 0402 parts and there is nowhere on the PCB close enough for a useful low value bypass... your process seemed to be just choosing a mixture of caps and putting them where they will fit, with no measurement; can you explain the similarity?
that is what I critiqued, not the use of the parallel parts. the technique is good, but 'risky' to do blind.
Last edited:
you may not hear them, but they are at a level that will meaningfully impact on the SNR/DNR/THD of the ESS analogue stages. yes I know the calc, but dont have the time, or desire to do it to illustrate a point for you.
I looked to see if I could find the stuff on the forum about regulator noise with no luck, but its from an in depth article on regulator noise in linear audio, which is a hard copy magazine/manual. its in Volume 4, the latest one; lots of other great stuff in there.
200 pages in this latest volume, check out the contributors, its a master class in several very well aligned subjects by some of the legends of the industry, the articles are VERY extensive. its cheap IMO, well not cheap, but its not disproportionate to the cost of putting together something like that in short run printed form.
Last edited:
This is a very interesting book.
The graphs for the regulator comparisons are difficult to read.
Here they are in colour
Online Articles
The graphs for the regulator comparisons are difficult to read.
Here they are in colour
Online Articles
thanks, I knew I had seen them somewhere in colour, still worth getting the hard copy IMO, because the article is excellent.
this latest edition is the best yet IMO, but every one ive read has had something very interesting and useful and there is always at least one free design/project. I think as a collection they will be a very good resource. I do wish there was a PDF/Ebook version, as the postage increases the price quite a bit to AU and takes a while.
the sjostrom super reg was a surprise, I have been using them for a couple years in a few projects and have liked them a lot, but I never expected them to be significantly better than the colony boards (for noise anyway)
but it really does show how well the salas reg works, particularly the impedance vs FS, so flat. I just wish they werent so massive and...well.. HOT!
the burson.... not much to say there....
this latest edition is the best yet IMO, but every one ive read has had something very interesting and useful and there is always at least one free design/project. I think as a collection they will be a very good resource. I do wish there was a PDF/Ebook version, as the postage increases the price quite a bit to AU and takes a while.
the sjostrom super reg was a surprise, I have been using them for a couple years in a few projects and have liked them a lot, but I never expected them to be significantly better than the colony boards (for noise anyway)
but it really does show how well the salas reg works, particularly the impedance vs FS, so flat. I just wish they werent so massive and...well.. HOT!
the burson.... not much to say there....
Last edited:
Thank you.
Those are noise density curves in nV/sqrt(Hz) units
A quick calculation shows that the bottom two curves represent 1 uV RMS and 1.4 uV RMS in the 10-100K frequency range.
Very impressive numbers. (but the commercial offerings are getting there. The 2-buck ADP-151 has 9uV RMS of noise in the 10-100KHz range)
Last edited:
yeah I think there is something wrong with your calculations (maybe not and i'm not about to do a back of the hand calc at 4am) and your comparison is pretty darn simplified. all of a sudden noise is the only variable.
YES noise is getting there with the best examples under lab conditions applied by the designers (why else am I also excited about the tps?), but is still 10x higher. for PSRR, Zout, bandwidth etc, LDOs are not even in the same ballpark.. PSRR in particular is of concern, so you better have a good prereg.
the 2900 would get utterly PWNED in the rest of the tests and anything that relies on the caps for transient response, PSRR and has low slew rate, will be more touchy with peaks and resonance and place even more pressure on the local decoupling.
they are getting there, but they are nowhere near there. 10-100K is a pretty biased bandwidth too, its common for LDOs to be tested that way, because they have crap performance above 100K amd are not usually expected to perform well higher due to leaning on the caps. the TPS7A will be better here but still limited.
how many ways can we say the same thing?
YES noise is getting there with the best examples under lab conditions applied by the designers (why else am I also excited about the tps?), but is still 10x higher. for PSRR, Zout, bandwidth etc, LDOs are not even in the same ballpark.. PSRR in particular is of concern, so you better have a good prereg.
the 2900 would get utterly PWNED in the rest of the tests and anything that relies on the caps for transient response, PSRR and has low slew rate, will be more touchy with peaks and resonance and place even more pressure on the local decoupling.
they are getting there, but they are nowhere near there. 10-100K is a pretty biased bandwidth too, its common for LDOs to be tested that way, because they have crap performance above 100K amd are not usually expected to perform well higher due to leaning on the caps. the TPS7A will be better here but still limited.
how many ways can we say the same thing?
Last edited:
without an MCU (or arduino) it wont accept spdif at all
but yes there is no multichannel spdif on the ESS, you cannot have a stereo channel going to D4 and another stereo channel to D6 and choose them to go to different dacs internally. you can only have multichannel PCM/i2s/DSD, spdif will only ever play one stereo stream and use 4 dacs each in stereo mode, or 8 dacs in mono
but yes there is no multichannel spdif on the ESS, you cannot have a stereo channel going to D4 and another stereo channel to D6 and choose them to go to different dacs internally. you can only have multichannel PCM/i2s/DSD, spdif will only ever play one stereo stream and use 4 dacs each in stereo mode, or 8 dacs in mono
Last edited:
Just trying to compare data with the same units... Either RMS or noise density... That's all. Same with jitter comparison (5 psec jitter needs the units: RMS, peak to peak, phase noise, period, etc)
Certainly the other parameters are just as important as well as intended application. No question about them...
Certainly the other parameters are just as important as well as intended application. No question about them...
In h/w mode, auto detection is only used on data 1. You must put SPDIF on DATA 1. If you want to put spdif on another input, you need a uP to tell it to use another data in
- Status
- Not open for further replies.