Options...
Hi ds23man,
I thanks for your suggestions re Cobranet and connecting remote DAC's /DSP, very interesting.
Hi Twest,
Thanks, good points. I like Mini DSP stuff but like the DEQX, the Mini DSP DAC's and power supply are not great. In order to run a full 3 way active set up you have to buy two of their DSP only units and you still lose out on performance and features compared to DEQX.
Its still better value than DEQX but not best performance and still mid market price.
I have been waiting a long time for Bruno to release his Hypex solution, he has the potential to really shake things up, I hope he does!
All the best
Derek.
Hi ds23man,
I thanks for your suggestions re Cobranet and connecting remote DAC's /DSP, very interesting.
Hi Twest,
Thanks, good points. I like Mini DSP stuff but like the DEQX, the Mini DSP DAC's and power supply are not great. In order to run a full 3 way active set up you have to buy two of their DSP only units and you still lose out on performance and features compared to DEQX.
Its still better value than DEQX but not best performance and still mid market price.
I have been waiting a long time for Bruno to release his Hypex solution, he has the potential to really shake things up, I hope he does!
All the best
Derek.
Derek, is your assessment of the miniDSP DACs and PSU by chance based on ABX or similar testing? Unfortunately I don't have access to either a 2x4 or 2x8 for ABX testing so I'm limited to speculation here but third party measurements of the 2x4 posted in the miniDSP forum here on DIYA are clean. I've never measured better than 70dB SNR in a recording environment and miniDSP is getting the ADAU1701's -90dB THD and 104dB DNR typ. So I'd expect the 2x4 to be fine with the usual 16 bit 44.1 material provided the downstream gain structure allows good bit depth utilization in the DAC---I could see a problem if using a power amp with the typical ~28dB gain for 50-60dB SPL listening levels but amps with a gain around unity ought to be OK if linear phase is one's thing (Analog doesn't spec the ADAU1701's group delay, which usually indicates linear phase antialiasing). For the 2x8 I haven't seen measurements but---if it is using Cirrus's -100dB THD/114dB DNR block---the only turnkey solutions I know of with higher headline specs and near minimum phase responses are the CS4398 (-107dB THD, 120dB DNR) and certain configurations of the WM8741 and 8742 (-100dB THD/123-125dB DNR). If one finds that extra bit is make or break it's probably not a bad idea to look into adjusting the playback channel's gain structure for better bit depth utilization in the DAC.
For example, I've gone through that exercise with CS4272s (one of Cirrus's -100dB/114dB DNR parts) and amps ranging from 0dB to 28dB gain (all of comparable THD, slew rate, and so on) and the results were kind of surprising. In the test configuration the ~27dB gain amps measured around -50dB acoustic THD for conversational listenling levels due to hitting the DAC floor. Switching to amps with around 10dB gain resolved the speakers -55 to -60dB THD but the ABX results from critical listening were 100% for the lower gain amps. Not by a small margin either; it was more like "why are we bothering to test this?"
As for power supplies, do you have specific concerns with the 2x8? (The 2x4 doesn't seem to be power supply limited.) The eval boards and reference designs for Cirrus's -100dB THD/114dB DNR all use a +5V lab supply with LC filters that corner around 3.5kHz. Statistically speaking that pretty much says it's an LM317 for the regulator but 3-4kHz is a typical of the regulation bandwidth of linear IC regulators. So the filtering is well chosen and should play nicely with pretty much any reasonable choice. That includes switchmode; the LC filters Cirrus uses hit peak rejection in the same range as the harmonics of the usual ~500kHz switchers. One might be able to wring a few dB more performance out of the DAC with some kind of super regulator but---given there's usually a few dB of performance variation from one chip to another due to process tolerances in the fab---the merits of that are debatable. Everything in Cirrus's design collateral suggests the performance limit is the DAC itself. Which is to be expected since manufacturers have quite a bit of incentive to set up their eval boards for the best headline specs possible.
I'm using a similar LC filter approach on my board with NCP1117s, which are ON's take on the LM1117. Great parts. The TDK MLZ inductors I'm using sim a little better than the ones Cirrus uses due to their lower ESR, too. But it's not until one gets everything assembled and ABXes it that one has a meaningful result. I am quite sure, however, that I'll get zero audiophile street cred from using a 48 cent regulator and 14 cent inductors even if the solution should be perfectly adequate to yield speced DAC performance---some exotic shunt regulator that didn't actually work as well would probably be perceived as better.
For example, I've gone through that exercise with CS4272s (one of Cirrus's -100dB/114dB DNR parts) and amps ranging from 0dB to 28dB gain (all of comparable THD, slew rate, and so on) and the results were kind of surprising. In the test configuration the ~27dB gain amps measured around -50dB acoustic THD for conversational listenling levels due to hitting the DAC floor. Switching to amps with around 10dB gain resolved the speakers -55 to -60dB THD but the ABX results from critical listening were 100% for the lower gain amps. Not by a small margin either; it was more like "why are we bothering to test this?"
As for power supplies, do you have specific concerns with the 2x8? (The 2x4 doesn't seem to be power supply limited.) The eval boards and reference designs for Cirrus's -100dB THD/114dB DNR all use a +5V lab supply with LC filters that corner around 3.5kHz. Statistically speaking that pretty much says it's an LM317 for the regulator but 3-4kHz is a typical of the regulation bandwidth of linear IC regulators. So the filtering is well chosen and should play nicely with pretty much any reasonable choice. That includes switchmode; the LC filters Cirrus uses hit peak rejection in the same range as the harmonics of the usual ~500kHz switchers. One might be able to wring a few dB more performance out of the DAC with some kind of super regulator but---given there's usually a few dB of performance variation from one chip to another due to process tolerances in the fab---the merits of that are debatable. Everything in Cirrus's design collateral suggests the performance limit is the DAC itself. Which is to be expected since manufacturers have quite a bit of incentive to set up their eval boards for the best headline specs possible.
I'm using a similar LC filter approach on my board with NCP1117s, which are ON's take on the LM1117. Great parts. The TDK MLZ inductors I'm using sim a little better than the ones Cirrus uses due to their lower ESR, too. But it's not until one gets everything assembled and ABXes it that one has a meaningful result. I am quite sure, however, that I'll get zero audiophile street cred from using a 48 cent regulator and 14 cent inductors even if the solution should be perfectly adequate to yield speced DAC performance---some exotic shunt regulator that didn't actually work as well would probably be perceived as better.
Five pages of text and I don't think that anyone has expressed an opinion
about what's the worlds best dsp chip yet. If you would like to contribute don't
mind the cost, availability and so on. My idea with this thread wasn't to find an
easy or affordable solution. Simple question, wish one would you choose no
matter what ?
Wikipedia gave some background (modern dsp's).
http://en.wikipedia.org/wiki/Digital_signal_processor
about what's the worlds best dsp chip yet. If you would like to contribute don't
mind the cost, availability and so on. My idea with this thread wasn't to find an
easy or affordable solution. Simple question, wish one would you choose no
matter what ?
Wikipedia gave some background (modern dsp's).
http://en.wikipedia.org/wiki/Digital_signal_processor
Last edited:
That's because it doesn't really matter. Microcontrollers tend to be a bit more performant for IIR and DSP architectures a bit more performant for FIR but pretty much any current chip supports Q31x63 and finding DSCs or DSPs with more than enough processing power for most things isn't at all difficult (though sometimes one ends up with a multi-chip solution). You'll find various preferences about various toolchains and hardware details---Abraxalito's open source DSP XO thread in this forum has quite a bit of discussion about that---and I've already described the core tradeoffs between ARM Cortex M4, SigmaDSP, SHARC, and Blackfin on this thread. Plus there's a great deal of discussion about PC crossovers in this forum, the PC based forum, and even the multi way forum.
As usual, there's no one best; it depends on the application one has in mind and what the design constraints are. Post 1 in this thread is, after all, one of those very broad questions that can be good for stimulating discussion but usually are not so good at producing specific answers. But, as you might guess from my last several posts in this thread, my choice of the moment for embedded applications is NXP's LPC4300 family. You can find details of my PC crossover here.
As usual, there's no one best; it depends on the application one has in mind and what the design constraints are. Post 1 in this thread is, after all, one of those very broad questions that can be good for stimulating discussion but usually are not so good at producing specific answers. But, as you might guess from my last several posts in this thread, my choice of the moment for embedded applications is NXP's LPC4300 family. You can find details of my PC crossover here.
picking the dsp is only one small part of the equation. You then have to design the hardware, layout the board, get boards made, build up prototypes and then write code for it
Even if you could use an off the shelf board that had everything you wanted on it you would still have to write the software. Again that is a major undertaking !!
Not for your average diyer I'm afraid.
Last edited:
just wondering here guys. all that talk about being on the same board, I didnt mention it at the time, but its worth the query. wouldnt it simply be better to hold the clock with the dac/s and isolate the DSP? it just seems that the DSP brings with it high slew rates, simultaneous switching noise and ripple, that to me has the potential to create more noise and limit the performance of the dac output more than a small amount of jitter from having the DSP on another module installed in the same case
have you done any testing on that twest820? it seems to me you would need to go to great trouble with multilayer boards to isolate the DSP, while going to efforts to keep it on the same board....
have you done any testing on that twest820? it seems to me you would need to go to great trouble with multilayer boards to isolate the DSP, while going to efforts to keep it on the same board....
Depends on the quality of the power supply implementation and the specifics of one's jitter concerns. Cirrus's LC approach I mentioned in post 42 is a good default. It's capable of providing 100+dB of isolation between analog and digital planes tied to the same regulator from a few kHz to a few hundred MHz.
Below 40kHz doesn't worry me much and that's where audible baseband jitter would occur. The lowest frequency noise source from the DSP that I know of is load pulsation with the sample rate---when a sample comes in the processor will be active for some duration depending on the amount of DSP to be done and then idle until the next sample arrives. In simple stereo processing implementations the fundamental of this supply loading is four times the Nyquist rate and therefore out of band pretty much by definition. Batch processing of samples due to loop unrolling or such can reduce it to frequencies below the Nyquist rate. But that's within or near regulator bandwidth and most parts have decent noise rejection in the kHz to a few tens of kHz range due to high PSRR and Avol.
From perhaps 100MHz up I'm less confident in what happens. Models get rather sensitive to what parasitics one decides to plug in and I don't have USD 50k lying around to buy a good scope with a few GHz of bandwidth to measure what's going on. It is, after all, a bit tricky to implement filters when everything's an inductor. Probably a decent rule of thumb is the supply filters need to be effective out to the GBP of the analog components so digital twitches don't blow into the op amps and burn all the slew rate and gain margin---in most cases the MCLK oscillator won't be any faster than the op amps so it's protected by this too. I'm using 50MHz typ op amps in the output buffers and I'm simming 120dB supply isolation at 300MHz so I figure things ought to be more than fine. Gotta build it and see, though.
Something to note about Cirrus's eval board schematics is the filter between planes is a CLCLC so supply noise sees two full LC stages no matter which direction it's moving in addition to the local bypassing for the chip creating the noise. A significant detail is each capacitor is implemented as an electrolytic in parallel with an MLCC. In the hundreds of MHz the MLCCs provide about 20dB of isolation each. If the layout's tight.
Below 40kHz doesn't worry me much and that's where audible baseband jitter would occur. The lowest frequency noise source from the DSP that I know of is load pulsation with the sample rate---when a sample comes in the processor will be active for some duration depending on the amount of DSP to be done and then idle until the next sample arrives. In simple stereo processing implementations the fundamental of this supply loading is four times the Nyquist rate and therefore out of band pretty much by definition. Batch processing of samples due to loop unrolling or such can reduce it to frequencies below the Nyquist rate. But that's within or near regulator bandwidth and most parts have decent noise rejection in the kHz to a few tens of kHz range due to high PSRR and Avol.
From perhaps 100MHz up I'm less confident in what happens. Models get rather sensitive to what parasitics one decides to plug in and I don't have USD 50k lying around to buy a good scope with a few GHz of bandwidth to measure what's going on. It is, after all, a bit tricky to implement filters when everything's an inductor. Probably a decent rule of thumb is the supply filters need to be effective out to the GBP of the analog components so digital twitches don't blow into the op amps and burn all the slew rate and gain margin---in most cases the MCLK oscillator won't be any faster than the op amps so it's protected by this too. I'm using 50MHz typ op amps in the output buffers and I'm simming 120dB supply isolation at 300MHz so I figure things ought to be more than fine. Gotta build it and see, though.
Something to note about Cirrus's eval board schematics is the filter between planes is a CLCLC so supply noise sees two full LC stages no matter which direction it's moving in addition to the local bypassing for the chip creating the noise. A significant detail is each capacitor is implemented as an electrolytic in parallel with an MLCC. In the hundreds of MHz the MLCCs provide about 20dB of isolation each. If the layout's tight.
but you still have shared ground, or am I missing something? check here for what i'm talking about in comparison. now this is a 2 channel version, something similar for multichannel is what I would call ideal. the reclock stage sits on the dac side, everything else including the connected PC and i2s buffer remains on the other side, so the only jitter that remains is that of the clock itself, the clock buffers, the interconnect and the dac itself.
its necessarily complicated and not cheap to implement, but effective. the difference between his and my setup currently is I have u.fl and w.fl connectors natively on the ackodac, so no adapters needed. this setup is confirmed bitperfect at 32bit and in my world this man knows how to make a PCB. the rest of the thread, or just the manuals linked in his sig will give you an indication of how the rest of the design works. with boards designed to work together it could be quite neat.
I know what you mean though, not saying its not possible, but there are certainly a fair few boxes to tick and its very difficult to really know whats going on at high speeds, as you say, due to the models becoming a bit bunky up there and measurement gear prohibitively expensive
its necessarily complicated and not cheap to implement, but effective. the difference between his and my setup currently is I have u.fl and w.fl connectors natively on the ackodac, so no adapters needed. this setup is confirmed bitperfect at 32bit and in my world this man knows how to make a PCB. the rest of the thread, or just the manuals linked in his sig will give you an indication of how the rest of the design works. with boards designed to work together it could be quite neat.
I know what you mean though, not saying its not possible, but there are certainly a fair few boxes to tick and its very difficult to really know whats going on at high speeds, as you say, due to the models becoming a bit bunky up there and measurement gear prohibitively expensive
well, ideal really would be if it was all on the one board, but I think I would still like to see the dac, i2s buffer and clock side side isolated from the DSP and PC. couldnt agree more about gain structure, particularly when you take the traditional crossover out of the question, its surprising how little gain one actually needs
you might look at the Murata EMIFIL CLCLC and CRCLCRC integrated filter parts
you might look at the Murata EMIFIL CLCLC and CRCLCRC integrated filter parts
Last edited:
Cirrus floods ground on both sides of the board and routes power in traces; have a look at appendix E of the CS42528+CS43900 reference design. The performance isn't speced but it's fairly safe to assume the board yields the codec's datasheet performance---reference designs which don't deliver this have a way of making large customers talk about cancelling contracts and switching to different vendors, so they tend to get fixed pretty quickly. The layout style is used consistently across the other Cirrus eval boards and reference designs I've looked at, including the CS4398, so it would appear to be rather not broken and therefore in no real need of fixing.
The Focusrite Saffire 40 I use in my PC crossover uses a similar approach to layout with CS4272s and NJM4565s. It's also a mixed signal system and is DC grounded to a laptop via 1394. It measures out at the CS4272's datasheet performance on loopback tests with a power supply that's a small EI trafo, two LM317s, and an LM337. I use it for recording as well as for VST hosted XO and EQ on playback and it does a fine job of satisfying the pro audio design basic of sound out = sound in. What limits sound quality is the recording which, in my case, is mostly constrained by the mics I have access to. I think I might able to expose some limitations of the system if I ever get ahold of a set of Earthworks QTCs but, comparing my own instruments to the best professional recordings I have in my collection, any technical limitations of the rig in playback are decidedly smaller than the variation in recording quality among, oh, I dunno, the top 10% of CDs in terms of production values. (I've also done what I can to ensure the speakers and amps are as high fidelity as I know how to get them; see link in post 44.)
As such, I feel I've a handle on what's good enough to achieve subjective transparency. At least in my living room for my ears, anyway. These subjective results are in agreement with analytic design workback; every time I end up running the maths I end up concluding one wants good time domain behavior and THD of -100dB or below and 100+dB DNR in each analog stage in order to keep the electronics' IMD level and noise floor low enough playback performance limits on the phase modulation distortion floor of the drivers. If one's aggressively managing levels the way you do on a mixing desk then one can get away with less performant gear. If one is underutilizing a power amp with a lot of gain (an unfortunately common problem) then probably more performant stuff will be helpful.
Cirrus's -100dB THD/114dB DNR lineup is nicely positioned for this. The parts are good enough. The parts are cheap. The antialising filters are well chosen. And you sweat the design work considerably less than if you're trying to hit the performance limits of one of ESS's -120dB THD/129+dB DNR DACs. Reducing a noise floor down into the nV range has a way of being brutal---I've discovered I really like not having to worry about how to manage Johnson noise from a 2k resistor---but if a few uV is OK then things become much more plug and play.
Thanks for the EMIFIL suggestion. I'd found the parts a while back, actually, and ended up rolling my own filters as the EMI market is aimed at frequencies rather higher that what one wants for audio (one or two of the parts come close but they're non-stock at Digikey). Something that gives me confidence in my sims is they show the same sort of behavior as Murata's BNX insertion loss graphs. I suspect the primary reason I'm getting significantly higher rejection is I'm designing for ~100mA rather than 10+A and can therefore use 0805 chip inductors. Those have lower parallel capacitances the inductors in the BNX parts and I have the freedom to chose higher ESRs than the BNX current ratings allow (there's something of a sweet spot around a couple hundred milliohms where the filtering's good and supply sag is minimal).
The Focusrite Saffire 40 I use in my PC crossover uses a similar approach to layout with CS4272s and NJM4565s. It's also a mixed signal system and is DC grounded to a laptop via 1394. It measures out at the CS4272's datasheet performance on loopback tests with a power supply that's a small EI trafo, two LM317s, and an LM337. I use it for recording as well as for VST hosted XO and EQ on playback and it does a fine job of satisfying the pro audio design basic of sound out = sound in. What limits sound quality is the recording which, in my case, is mostly constrained by the mics I have access to. I think I might able to expose some limitations of the system if I ever get ahold of a set of Earthworks QTCs but, comparing my own instruments to the best professional recordings I have in my collection, any technical limitations of the rig in playback are decidedly smaller than the variation in recording quality among, oh, I dunno, the top 10% of CDs in terms of production values. (I've also done what I can to ensure the speakers and amps are as high fidelity as I know how to get them; see link in post 44.)
As such, I feel I've a handle on what's good enough to achieve subjective transparency. At least in my living room for my ears, anyway.
These subjective results are in agreement with analytic design workback; every time I end up running the maths I end up concluding one wants good time domain behavior and THD of -100dB or below and 100+dB DNR in each analog stage in order to keep the electronics' IMD level and noise floor low enough playback performance limits on the phase modulation distortion floor of the drivers. If one's aggressively managing levels the way you do on a mixing desk then one can get away with less performant gear. If one is underutilizing a power amp with a lot of gain (an unfortunately common problem) then probably more performant stuff will be helpful.Cirrus's -100dB THD/114dB DNR lineup is nicely positioned for this. The parts are good enough. The parts are cheap. The antialising filters are well chosen. And you sweat the design work considerably less than if you're trying to hit the performance limits of one of ESS's -120dB THD/129+dB DNR DACs. Reducing a noise floor down into the nV range has a way of being brutal---I've discovered I really like not having to worry about how to manage Johnson noise from a 2k resistor---but if a few uV is OK then things become much more plug and play.
Thanks for the EMIFIL suggestion. I'd found the parts a while back, actually, and ended up rolling my own filters as the EMI market is aimed at frequencies rather higher that what one wants for audio (one or two of the parts come close but they're non-stock at Digikey). Something that gives me confidence in my sims is they show the same sort of behavior as Murata's BNX insertion loss graphs. I suspect the primary reason I'm getting significantly higher rejection is I'm designing for ~100mA rather than 10+A and can therefore use 0805 chip inductors. Those have lower parallel capacitances the inductors in the BNX parts and I have the freedom to chose higher ESRs than the BNX current ratings allow (there's something of a sweet spot around a couple hundred milliohms where the filtering's good and supply sag is minimal).
Member
Joined 2009
I'm sorry to hijack the thread but I need a quick advice and you guys seem very knowledgeable on the subject. I just bought a miniDSP and while evaluating it I discovered that it's very sensitive to different power supplies. (Contrary to the manifacturer's claim)
I'm currently using it with this one but I find the noise floor to be a bit high and the bass transients are still garbled. Can you guys point me to commercial solutions that could improve the miniDSP performance or do you consider it a bad design that can't be improved with a power supply?
What are desirable characteristics for a good power supply?
I'm currently using it with this one but I find the noise floor to be a bit high and the bass transients are still garbled. Can you guys point me to commercial solutions that could improve the miniDSP performance or do you consider it a bad design that can't be improved with a power supply?
What are desirable characteristics for a good power supply?
Hi Boris, I'd say your core question would be best asked in the miniDSP forum; more likely that other users with the same issue will see it, plus miniDSP support may respond.
Posts 46 to 50 discuss one method of answering your secondary question. If you can figure out how miniDSP implements the supply topology on their board and post a schematic, board layout pics, and measurements here that would be an quite interesting data point to bring up alongside Cirrus's and Focusrite's design approaches. Sorry if this sounds picky, but it's hard to say much that's meaningful without some understanding of the design context. And, well, post 51 doesn't indicate which miniDSP units are in use much less resolve the guesses earlier in this thread as to what parts miniDSP is using.
DACs generally have around 50dB PSRR typ so one tends to need to hold analog supply ripple below a couple hundred microvolts to expose the DAC's DNR as the noise floor at the low PSRR corner. That's doable with regular parts but wants some attention to detail to pull off as it's demanding of regulators' own PSRR and noise floors---if noise performance beyond the usual LM78xx, LM317, LM1086, and LM1117 parts is needed the ADP150 and 151 are good starting points. For switchmode wall warts you pretty much have to measure to find out the switching frequency and ripple and then design from there as exceedingly few of them are actually specified. If you control the part choice it's rather easier; some of the parts in National's SimpleSwitcher line deliver a few millivolts of ripple at 500kHz, which is both small and easily rejected by audio band filters. However, DAC current draws are generally low enough switchmode isn't particularly cost effective compared to linear when one's DIYing the supply.
As an aside, ESS doesn't spec their DACs' PSRR but Dustin's indicated here on DIYA he designed for performance at the expense of PSRR so it's likely their PSRR is noticeably lower. That's likely why their parts require low noise, high PSRR, cap swamped op amps as regulators to hit the datasheet specs.
Posts 46 to 50 discuss one method of answering your secondary question. If you can figure out how miniDSP implements the supply topology on their board and post a schematic, board layout pics, and measurements here that would be an quite interesting data point to bring up alongside Cirrus's and Focusrite's design approaches. Sorry if this sounds picky, but it's hard to say much that's meaningful without some understanding of the design context. And, well, post 51 doesn't indicate which miniDSP units are in use much less resolve the guesses earlier in this thread as to what parts miniDSP is using.
DACs generally have around 50dB PSRR typ so one tends to need to hold analog supply ripple below a couple hundred microvolts to expose the DAC's DNR as the noise floor at the low PSRR corner. That's doable with regular parts but wants some attention to detail to pull off as it's demanding of regulators' own PSRR and noise floors---if noise performance beyond the usual LM78xx, LM317, LM1086, and LM1117 parts is needed the ADP150 and 151 are good starting points. For switchmode wall warts you pretty much have to measure to find out the switching frequency and ripple and then design from there as exceedingly few of them are actually specified. If you control the part choice it's rather easier; some of the parts in National's SimpleSwitcher line deliver a few millivolts of ripple at 500kHz, which is both small and easily rejected by audio band filters. However, DAC current draws are generally low enough switchmode isn't particularly cost effective compared to linear when one's DIYing the supply.
As an aside, ESS doesn't spec their DACs' PSRR but Dustin's indicated here on DIYA he designed for performance at the expense of PSRR so it's likely their PSRR is noticeably lower. That's likely why their parts require low noise, high PSRR, cap swamped op amps as regulators to hit the datasheet specs.
Last edited:
Member
Joined 2009
Twest, thank you for the quick response. The topics you are discussing are beyond my level of understanding so I doubt I can contribute much. The guys at the miniDSP forum suggested I evaluate it using a battery and now I can confirm that it performs very well (subjectively). I think the measurements you mentioned in post# 42 are genuine but the trick is finding a power supply that performs similar to a battery.
The link you gave was to a switching supply - they do tend to contribute high levels of common-mode noise which is very hard to filter out. The battery has zero common-mode noise as its isolated. In between these two would be a linear power supply, using an EI core transforrmer - this has better isolation than a toroidal transformer but much worse than a battery.
Yeah, the 0dB CMRR of unbalanced interconnects is one of the home audio's industry's more notable disservices to its customers. All it takes to build a differential receiver is two extra resistors and, since the default resistor tolerance is 1% at the moment, that's 40dB CMRR. Susumu's RR series is nice---0.5% for at most 1 cent more per resistor than 1% and pretty low temp co---and it's tough to find better than the RGs if one needs more than the 46dB CMRR the 0.5% RRs provide.
Most times I run the maths 40dB is sufficient to render the likely range of ground bounce between components inaudible if the gain structure of the system is well configured. But the extra 6dB from the RRs is cheap insurance. At the other extreme a THX certified power amp (28dB gain required) with unbalanced interconnects amplifies any bounce on its ground relative to the preamp by 28dB. With a typical AWG 16 power cord and the relatively large quiescent dissipation to output signal power of a typical 100-200W solid state amp used at conversational listening levels it actually works out the ground bounce created across the power cord's resistance by the charging pulses through the amp's rectifier is amplified enough that it's borderline audible. The details depend on the amp's ground structure as well as what's going on with the preamp/DAC's own charging pulses and the bounce is often down around a millivolt where it's tricky to measure. (Lest folks think I'm off my rocker here's a basic run through of the math: output levels for typical listening tend to be a few hundred mV and the edge of audibility is nominally around 50dB below signal so, with the amp's 28dB gain, the ground voltage across an unbalanced interconnect needs to be tens of microvolts or less to be inaudible. AWG 16 wire is 13 milliohms per meter and power amps generally pull around an amp in normal operation so the topology is not set up for success; it's aimed more at tens of millivolts instead of the tens of microvolts needed.)
Not clear from the supply thread in the miniDSP forum if Boris has the balanced or unbalanced version of the 2x4, though. Boris, which version do you have?
If one's DC coupling a DSP and DAC this problem gets a little trickier as the digital source may be USB or 1394, either of which may bring more ground noise than a wired SPDIF connection to CD player or a Squeezebox would. The circuit topology doesn't change, though; one still has a source, a DSP/XO/EQ/DAC/preamp thingy, and a power amp. They all have their own power cords, so there are three different loops for ground currents to run around in. Since it's all low impedance and the ground net is more or less everywhere the voltages are low and it takes good (read more expensive than most DIY budgets) measurement gear to get measurements good enough to have clarity on what's going on. Subjectively I think I can sometimes hear a small difference with the laptop and Saffire plugged into different wall outlets on the same circuit versus the same power strip, which would imply a kind of disturbing amount of noise in the ground loop from the laptop supply, but I've never been able to get a statistically signifcant ABX result out of it.
Most times I run the maths 40dB is sufficient to render the likely range of ground bounce between components inaudible if the gain structure of the system is well configured. But the extra 6dB from the RRs is cheap insurance. At the other extreme a THX certified power amp (28dB gain required) with unbalanced interconnects amplifies any bounce on its ground relative to the preamp by 28dB. With a typical AWG 16 power cord and the relatively large quiescent dissipation to output signal power of a typical 100-200W solid state amp used at conversational listening levels it actually works out the ground bounce created across the power cord's resistance by the charging pulses through the amp's rectifier is amplified enough that it's borderline audible. The details depend on the amp's ground structure as well as what's going on with the preamp/DAC's own charging pulses and the bounce is often down around a millivolt where it's tricky to measure. (Lest folks think I'm off my rocker here's a basic run through of the math: output levels for typical listening tend to be a few hundred mV and the edge of audibility is nominally around 50dB below signal so, with the amp's 28dB gain, the ground voltage across an unbalanced interconnect needs to be tens of microvolts or less to be inaudible. AWG 16 wire is 13 milliohms per meter and power amps generally pull around an amp in normal operation so the topology is not set up for success; it's aimed more at tens of millivolts instead of the tens of microvolts needed.)
Not clear from the supply thread in the miniDSP forum if Boris has the balanced or unbalanced version of the 2x4, though. Boris, which version do you have?
If one's DC coupling a DSP and DAC this problem gets a little trickier as the digital source may be USB or 1394, either of which may bring more ground noise than a wired SPDIF connection to CD player or a Squeezebox would. The circuit topology doesn't change, though; one still has a source, a DSP/XO/EQ/DAC/preamp thingy, and a power amp. They all have their own power cords, so there are three different loops for ground currents to run around in. Since it's all low impedance and the ground net is more or less everywhere the voltages are low and it takes good (read more expensive than most DIY budgets) measurement gear to get measurements good enough to have clarity on what's going on. Subjectively I think I can sometimes hear a small difference with the laptop and Saffire plugged into different wall outlets on the same circuit versus the same power strip, which would imply a kind of disturbing amount of noise in the ground loop from the laptop supply, but I've never been able to get a statistically signifcant ABX result out of it.
right, so its changed from a digital input, digital output DSP/XO thread, to being hijacked as an all in one analogue out digital XO thread, to a mini DSP power supply support thread?
sorry I just find that really frustrating! I was really happy to see this thread started because I was going to open one of my own on the same topic, then I got a bit pissed when it was taken over by people claiming the single interconnect for clock meant the thing was unworkable, thus better off building or buying an entire design I dont need, which was the entire point of wanting to start the thread about not doing that. NOT having to buy and bypass the dac and analogue stages was the entire point and nobody will convince me otherwise, now its gone somewhat further back on topic, but not really.
so do I start another thread called high end ALL DIGITAL DSP crossover, no DACs allowed?
sorry I just find that really frustrating! I was really happy to see this thread started because I was going to open one of my own on the same topic, then I got a bit pissed when it was taken over by people claiming the single interconnect for clock meant the thing was unworkable, thus better off building or buying an entire design I dont need, which was the entire point of wanting to start the thread about not doing that. NOT having to buy and bypass the dac and analogue stages was the entire point and nobody will convince me otherwise, now its gone somewhat further back on topic, but not really.
so do I start another thread called high end ALL DIGITAL DSP crossover, no DACs allowed?
Last edited:
I thought it had been hijacked until I read twest's last post, which I found completely on topic
A 'best' DSP however it is implemented and whether all digital or whether it contains analogue interfacing, is dependent on its interconnection and interaction with upstream and downstream components, and the quality of its supporting circuitry and detailed implementation (stating the obvious)
Back to the OP. Some of the respondents here consider 'state of the art' and 'all digital' not to be completely compatible, and it's therefore a fair response if the answers err away from 'all digital' and they explain why
Similarly, the answers so far would indicate that no-one is building a 'state of the art all digital' - so does that mean the thread is or should be closed?
Instead we have seen references to the closest anyone seems to be doing to the OP's spec whether or not they are 'all digital' and whether or not they actually exist yet - that seems fine to me too. And since such a device doesn't seem to exist, then observations on implementation - eg. balanced connections and psu considerations also seem relevant to me
Anyway, let's allow any relevant/on-topic posts to continue
This thread has been of topic from the beginning, it has been only about how to transport the digital stream to and from the DSP board.
But the main question is: what is a state of the art DSP?
So we have to get some answers on the next questions:
1. What is the best DSP chip out there?
2. What is the influence of the software running the chip?
I/O problems can be solved, but how do we "define" the state of the art DSP!
But the main question is: what is a state of the art DSP?
So we have to get some answers on the next questions:
1. What is the best DSP chip out there?
2. What is the influence of the software running the chip?
I/O problems can be solved, but how do we "define" the state of the art DSP!
ChrisPa, no respondents vs hijacked from the beginning are 2 completely different things
twests last post was somewhat more on topic, but still not really on topic, digital in-out noise floor and psu quality for that is pretty different to noise floor and gain structure in the analogue stages are they not? the analogue noise floor being carried through from the digital source and whether to isolate or not is relevant, but I could just see the next post being another mini DSP support question. just because there are few products in this category doesnt mean there shouldnt be a thread, this is DIYAUDIO isnt it?
i'm just cooking dinner, so will post something a bit more relevant in the next couple hours.
twests last post was somewhat more on topic, but still not really on topic, digital in-out noise floor and psu quality for that is pretty different to noise floor and gain structure in the analogue stages are they not? the analogue noise floor being carried through from the digital source and whether to isolate or not is relevant, but I could just see the next post being another mini DSP support question. just because there are few products in this category doesnt mean there shouldnt be a thread, this is DIYAUDIO isnt it?
i'm just cooking dinner, so will post something a bit more relevant in the next couple hours.
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.