Opamps as in Gain stages? - well maybe I'm getting older and deaf, but my experience of switching opamps is while you gain a little here - but loose a little elsewhere - like capacitors, I've yet to hear one that was totally transparent, its more of a matter of finding the best balance - never really heard one that I would be happy with.
However, there is a massive step change when going to a discrete design - again you might get more from a discrete design if you spend time tweaking, but even if time does not allow you gained so much over the IC opamp that you don't feel that you cheated the customer... Anything else then a discrete design is just "Cheap"... "Plastic"... when I have to use Opamps I feel like a "cheap bitch" as I know I'm not getting the best from the design - I'm letting myself and the customer down...
I can speak from recent experience as I'm currently working on a PCB where an IC opamp stage has been replaced by a discrete design.
I will say that I strongly believe that many of the DiY opamp rolling results in upgrades with opamps that are oscillating at RF. When I see say a LM5532 replace with a newer high speed opamps, they will nearly all have some sign of oscillation - I continually face this issue when working with higher speed devices.. it would account for the heat / power dissipation and brightness often reported..
Its also common to see the oscillation at say 500MHz and a few mV making it very hard to notice directly unless your looking for it - normally you will notice the THD increase / change a tad when probed with a scope probe, or the broadband noise floor is elevated a few dB higher then you expect...
Capacitors are just a huge bag of hurt - just stay away from ceramics... On my discrete DAC designs I need to use Ceramics on the PSU rails as I integrate high speed pulses (sadly I only know of ceramics that work at HF/RF), you can really hear them and I have to pull tricks with the regulators circuits to mitigate the effect of these ceramics on SQ... but one feels that the sound of the ceramics caps its still present to a varying (but lessor) degree...
Really avoid them at all costs in filtering / EQ and gain stage compensation networks!!!
Last edited:
Interesting. How (what are the tricks you use?) can one mitigate the negative effects of ceramic capacitors?
Recently I got rather depressed as I need to use ceramics capacitors to HF / RF decouple a discrete DAC array design to avoid nasty harmonics at lower signal levels (say -60dB), while decoupling the array PSU with ceramics resolved the measurement issue, but you could now really hear the hardness / brightness of the ceramics PSU decoupling capacitors...
Originally on the design, due to limited PCB area I used the ES9311 dual low noise regulator IC and used a network analyser to insure there was no peaking within the audio band etc. but with the ceramics it sound bright / hard / horrid... on the FFT, Noise floor was very low (Below -165dB ref 5V from about 100Hz upwards), with no humps within the noise floor to indicate any peaking etc.. but still very depressing SQ...
The solution was a CCS / Shunt regulator + the ceramics, this INSTANTLY resolved the brightness / hardness over a simple "Source" regulator such as the ES9311
I suspect that the reason that it works so well is that the Ceramic capacitors across the Audio B/W are Shunted by the regulators shunt element... The challenge was to design a circuit that still had gain across the entire audio B/W so that the shunt element was "active / responsive" even at 20KHz... The regulator still had loop-gain at 20KHz even with a bank of 4000uF of total decoupling capacitors on the DAC array PSU rails...
This required designing a simple discrete Gain stage for the regulator circuit - the one of many headaches is the 5.6V input rails for 5.2V output to the Array... leaving very little PSU headroom - and meeting the High PSRR and low noise requirement that meant paralleling many input devices etc... One has to admire the designer of the ES9311 - it has amazing technical performance (especially for an CMOS LDO)- a very clever design indeed
...In the end due to the lack of PCB space the now larger discrete regulator design has had to be relocated onto a separated PCB, with 4 wire Kelvin sensed feedback taken from the mid point of the DAC array... How the ES9311 had been such a neater - simpler solution had it sounded any good!!! - the only point (apart from the all important improvement in SQ) is that discrete circuits are cheaper then the ESS part - but then I need to factor the extra PCB area and board to board connectors, much greater design complexity etc...
In absolute terms I suspect I can still hear the ceramics, but it now does so much right that its just the price to pay until I can find something better.. if ever...
Measurement wise I see very little difference, the ES9311 has an odd LF wonder, I suspect due to some kind of LF intergartor used internally to filter the noise of its Band Gap reference - its output voltage wonders a few mV's slowly up and down on a say 10 second timebase...
The 4 wire Kelvin sensed feedback of the discrete helps to reduce the PSU impedance SEEN AT the DAC array - I see PSU rail Second harmonics (when measuring directly on the PSU rail) under full load at about -145dB instead of say -140dB with the ESS... I hardly suspect that this improvement of say 5dB is the cause of any sonic improvement (at such low levels)... Still its good to do the job properly... and really required in this case as the regulators are now remotely located (due to lack of PCB space for the larger discrete design).
I try to maintain atleast 65dB phase margin with the discrete regulator design - the ES9311 is not so high.. No difference is observed in audio output THD / SNR measurements between the two, but the dynamic performance of regulator circuits is an example of a parameter that cannot be directly measured on the audio output, but DOES very much impact the SQ IME.
Last edited:
Many off topic posts have been split Opamps as in Gain stages? - well maybe I'm getting older and deaf, but my experience of switching opamps is while you gain a little here - but loose a little elsewhere - like capacitors, I've yet to hear one that was totally transparent, its more of a matter of finding the best balance - never really heard one that I would be happy with.
However, there is a massive step change when going to a discrete design - again you might get more from a discrete design if you spend time tweaking, but even if time does not allow you gained so much over the IC opamp that you don't feel that you cheated the customer... Anything else then a discrete design is just "Cheap"... "Plastic"... when I have to use Opamps I feel like a "cheap bitch" as I know I'm not getting the best from the design - I'm letting myself and the customer down...
I can speak from recent experience as I'm currently working on a PCB where an IC opamp stage has been replaced by a discrete design.
I will say that I strongly believe that many of the DiY opamp rolling results in upgrades with opamps that are oscillating at RF. When I see say a LM5532 replace with a newer high speed opamps, they will nearly all have some sign of oscillation - I continually face this issue when working with higher speed devices.. it would account for the heat / power dissipation and brightness often reported..
Its also common to see the oscillation at say 500MHz and a few mV making it very hard to notice directly unless your looking for it - normally you will notice the THD increase / change a tad when probed with a scope probe, or the broadband noise floor is elevated a few dB higher then you expect...
Capacitors are just a huge bag of hurt - just stay away from ceramics... On my discrete DAC designs I need to use Ceramics on the PSU rails as I integrate high speed pulses (sadly I only know of ceramics that work at HF/RF), you can really hear them and I have to pull tricks with the regulators circuits to mitigate the effect of these ceramics on SQ... but one feels that the sound of the ceramics caps its still present to a varying (but lessor) degree...
Really avoid them at all costs in filtering / EQ and gain stage compensation networks!!!
Very interesting. Thanks for your valuable insight!
Recently I got rather depressed as I need to use ceramics capacitors to HF / RF decouple a discrete DAC array design to avoid nasty harmonics at lower signal levels (say -60dB), while decoupling the array PSU with ceramics resolved the measurement issue, but you could now really hear the hardness / brightness of the ceramics PSU decoupling capacitors...
Originally on the design, due to limited PCB area I used the ES9311 dual low noise regulator IC and used a network analyser to insure there was no peaking within the audio band etc. but with the ceramics it sound bright / hard / horrid... on the FFT, Noise floor was very low (Below -165dB ref 5V from about 100Hz upwards), with no humps within the noise floor to indicate any peaking etc.. but still very depressing SQ...
The solution was a CCS / Shunt regulator + the ceramics, this INSTANTLY resolved the brightness / hardness over a simple "Source" regulator such as the ES9311
The challenge was to design a circuit that still had gain across the entire audio B/W so that the shunt element was "active / responsive" even at 20KHz... The regulator still had loop-gain at 20KHz even with a bank of 4000uF of total decoupling capacitors on the DAC array PSU rails...
This required designing a simple discrete Gain stage for the regulator circuit - the one of many headaches is the 5.6V input rails for 5.2V output to the Array... leaving very little PSU headroom - and meeting the High PSRR and low noise requirement that meant paralleling many input devices etc... One has to admire the designer of the ES9311 - it has amazing technical performance (especially for an CMOS LDO)- a very clever design indeed
In the end due to the lack of PCB space the now larger discrete regulator design has had to be relocated onto a separated PCB, with 4 wire Kelvin sensed feedback taken from the mid point of the DAC array... How the ES9311 had been such a neater - simpler solution had it sounded any good!!! - the only point (apart from the all important improvement in SQ) is that discrete circuits are cheaper then the ESS part - but then I need to factor the extra PCB area and board to board connectors, much greater design complexity etc...
In absolute terms I suspect I can still hear the ceramics, but it now does so much right that its just the price to pay until I can find something better.. if ever...
Measurement wise I see very little difference, the ES9311 has an odd LF wonder, I suspect due to some kind of LF intergartor used internally to filter the noise of its Band Gap reference - its output voltage wonders a few mV's slowly up and down on a say 10 second timebase...
The 4 wire Kelvin sensed feedback of the discrete helps to reduce the PSU impedance SEEN AT the DAC array - I see PSU rail Second harmonics (when measuring directly on the PSU rail) under full load at about -145dB instead of say -140dB with the ESS... I hardly suspect that this improvement of say 5dB is the cause of any sonic improvement (at such low levels)... Still its good to do the job properly... and really required in this case as the regulators are now remotely located (due to lack of PCB space for the larger discrete design).
I try to maintain atleast 65dB phase margin with the discrete regulator design - the ES9311 is not so high.. No difference is observed in audio output THD / SNR measurements between the two, but the dynamic performance of regulator circuits is an example of a parameter that cannot be directly measured on the audio output, but DOES very much impact the SQ IME.
One question John. When you make these difficult design choices, do you always consider how the final product will be judged by measurements on sites like ASR, or would you be prepared to deliver a dac that measures somewhat mediocre when your ears tell you it sounds better that way?
Lucas,
Such a great question, the "Power" that sites like ASR are gaining is really concerning and detrimental to the end sound quality IME.
The trouble with the guys on ASR is that they ONLY judge a products performance via "simplistic" technical measurements and IME this does a massive disservice to the general HiFi community - especially those who are scared to listen for themselves...
Take for example the little ProJect S2 Prebox DAC (which is still one of the best selling DAC's at its price point), I was essentially forced by my "Concern" of such technophiles as ASR to include user selectable modes - Best measurement and Best quality...
Having such a mode on one of my designs, I get decent feedback from a large number of users and apart from the predictable "Narrow Minded" Technophiles I know of no one who prefers the Measurement mode for listening...
On a new design that will be GBP100 or so more expensive, I fear that ASR will rip it apart because in some ares it measures worst (despite being more expensive) - but I've engineered this design, understand the compromise and am NEVER going to be bullied into releasing a worst sounding product just to satisfy such a stupid narrow minded perspective.
I'm not advocating a bad design, when I feel a certain parameter is important then I'll never compromise, but I'll base any design compromises on a lifetime of experience in my field - its this lifetime of working knowledge that creates the value and respect of "my" designs...
As far as I'm concerned, in the case above, if you want "Best measurements", then buy the cheaper unit and be Happy and "Laugh" at the rest of us
but DON"T dare say the new design sounds worst because in some areas it has poorer measured performance...You will be surprised how many (it not all) designer I know will privately agree that standard "static" measurements have little refection on sound quality (try as they might to measure and "quantify" sound quality) - but just as in so many walks of life are too afraid to have there voices counted... Life is just so much easier if you don't place yourself in the firing line... and it can get VERY vicious very quickly!!!
Very true, unfortunately. Sorry to see good and honest designers treated so badly. Also sorry to see some unhappy consumers who feel that they were cheated. Doubtful a solution exists. Presumably, one could market to the people who don't care about anything but sound quality and who don't care too much what it costs to do it well. Its apparently a small volume market though.
Most people seem to want a lot for a little, not realizing for a little the best one gets is some nice looking numbers, and not especially good sound quality. That's okay with them though, since it suits their needs to believe they are getting a 'deal.'
Hi John. I'm just curious about when you don't want to use ceramics, what is your preferred alternate choice? When it comes to PSU filtering, the tiny capacitance values of film capacitors (per volume) make them pretty tough to choose from so I'm curious how you work around it. Thanks!
Michael
Lucas,
Such a great question, the "Power" that sites like ASR are gaining is really concerning and detrimental to the end sound quality IME.
The trouble with the guys on ASR is that they ONLY judge a products performance via "simplistic" technical measurements and IME this does a massive disservice to the general HiFi community - especially those who are scared to listen for themselves...
Take for example the little ProJect S2 Prebox DAC (which is still one of the best selling DAC's at its price point), I was essentially forced by my "Concern" of such technophiles as ASR to include user selectable modes - Best measurement and Best quality...
Having such a mode on one of my designs, I get decent feedback from a large number of users and apart from the predictable "Narrow Minded" Technophiles I know of no one who prefers the Measurement mode for listening...
On a new design that will be GBP100 or so more expensive, I fear that ASR will rip it apart because in some ares it measures worst (despite being more expensive) - but I've engineered this design, understand the compromise and am NEVER going to be bullied into releasing a worst sounding product just to satisfy such a stupid narrow minded perspective.
I'm not advocating a bad design, when I feel a certain parameter is important then I'll never compromise, but I'll base any design compromises on a lifetime of experience in my field - its this lifetime of working knowledge that creates the value and respect of "my" designs...
As far as I'm concerned, in the case above, if you want "Best measurements", then buy the cheaper unit and be Happy and "Laugh" at the rest of us
You will be surprised how many (it not all) designer I know will privately agree that standard "static" measurements have little refection on sound quality (try as they might to measure and "quantify" sound quality) - but just as in so many walks of life are too afraid to have there voices counted... Life is just so much easier if you don't place yourself in the firing line... and it can get VERY vicious very quickly!!!
Thanks John for your open and honest reply. I guessed as much, and I am glad I don't have to make these difficult decisions when building, modding or tweaking my own electronical stuff.
My own background is recording classical music, and many of my colleagues have long since discovered that the best measuring ADCs and DACs, are not the best sounding ones. Our work is 100% focussed on listening to the musicians and to what comes out of our monitors, so most of us have no scruples to trust our ears more than the data-sheets.
I do a lot of design on loudspeakers however, and there we have similar problems. I know some designers who have an excellent technical background, and who are really knowledgeable people, but they adhere to some rigid and simplified rules, like getting a flat freq. response with their measuring mics on a 1 meter distance. That has made a lot of their designs unlistenable since the way those waves reach our eardrums in an acoustical environment and in a longer distance makes a huge difference. The perceived tonal balance therefore is way off from what we are accustomed to when listening to a real performance and these speakers are therefore very tiring and annoying to listen to. When one brings up this subject to these designers, they simply dismiss any such notion because they think that without these well-established measurements practices, we would become a bunch of chaotic voodoo worshippers, and a well-educated scientist prefers to leave all psycho-acoustic considerations outside. That is the unfortunate reason for so many loudspeaker designs sounding completely unnatural in real-world applications.
That is not to say that measurements are not that important. Without measurements I would be truly lost, and they help me enormously with getting better enclosures, baffle design, filter design etc.… However, I do give my ears the final saying and when it’s clear to me there is a firm discrepancy between my measurement and my subjective impression, I will try to rethink the way how I do these measurements. Often that leads to new ways of measuring, but sometimes (frustratingly) I can’t yet explain these discrepancies and I simply have learned to live with that. One can be sorry for that, but it also presents us with an interesting and ever tempting future goal of getting a more comprehensive understanding of what is really going on..
Yes, you're right. It is considered a field that is outside their scope...
Its the AKM AK4499 evaluation board. I've been hacking it some. Was getting ready to start laying out a couple of PCBs for experiments to finish up settling on the front end part of a dac design.
Its sorta like I'm trying to learn some years of experience in shorter time. The best I can do requires running various experiments to see what makes the device sound good. Found out some interesting stuff, so the experiments have been productive if only slowly progressing. Would be happy to talk more offline.
Of course, it could be useful to understand more about what is going on inside the chip, like for instance why the eval board uses 7805 regulators for each channel's VDDxx supply, rather than a regulator more suited to digital loads. Must be something analog in there, but not as sensitive to voltage regulation as the VREFxx supply for each channel (which uses Jung regulators, presumably to power resistor networks).
Michael,
There's no single solution that fits all applications. Where larger capacitance values are required (for PSU decoupling) I use Organic / Polymer capacitors and supplement with Ceramics for Higher Frequency.
Organic capacitors seem to have a very good working life - I've yet to see one fail!
The discrete DAC array uses digital circuitry in an "Analogue" mode and requires decoupling at 100MHz+ this necessitates the need for large value ceramic capacitors - with the sonic struggles I described earlier...
I'v been reading though the Datasheet and running some Sims and I think I now have a better idea of whats going on with there architecture.
Just like the AK4137 Datasheet its really confusing and seemingly contradicts itself in so many places (our maybe for want of a better explanation).
For starters, I guess to clear thermal noise they run at crazy high output levels, almost 5Vrms Single ended, and 9.2V balanced!!! - to be fair they are not alone with ESS also pulling the same trick to be able to quote such high dynamic range figures on the datasheet...
The AK4499 is about the same level as the ES9028Pro in Mono mode, with a differential source impedance of 55 ohms..
A few points I'm struggling with - in I/V mode I don't see the advantage of the I/V stage feedback being returned to the die..
Also the cryptic line:-
"The output range of I-V conversion is 4.6 Vrms centered around signal common voltage, and 9.2 Vrms after differential summing. IOUTL1P/R1P/L2P/R2P current and IOUTL1N/R1N/L2N/R2N current cannot be summed. The differential outputs are summed externally after I-V conversion. "
Why cannot the currents be summed before I/V? maybe just due to the current requirement from the Opamp? - but a decent discrete stage would have little problem... Having paralleled I/V stages IME is just poor design - ESS also resorted to the same with there Larger 90x8Pro Devices
Last edited:
John,
It is possible to parallel directly. In practice.
Here it is an example, with transformer output..
やなさん AK4499DM DAC 音出し成功 : Studio Okamoto の 徒然日記
It is possible to parallel directly. In practice.
Here it is an example, with transformer output..
やなさん AK4499DM DAC 音出し成功 : Studio Okamoto の 徒然日記
Joseph,
Thank you - If the outputs could not be paralleled then I would struggled with understanding the internal architecture of the DAC array...
Mind you there are no measurements in the link so I hope its not a case of "Can work" - but not well...
There is something in the data sheet in section 10.4.3. Feedback Loop of External Operational Amplifier which suggests the I/V loop back is to keep the opamp output stable and within the voltage the dac chip can tolerate when the system is in the reset state. (Reset should be the state when the system is powered up and powered down, of course.)
There is some more at the bottom of page 88 which talks about possible damage to the chip and or muting of pops at power up/down.
I thought the I/V feedback routing might have to do with above sorts of concerns, but they never directly say so in the data sheet.
They did include reverse Schottky diodes on the eval board to prevent the opamp outputs from going below ground (on the filtered side of VREF).
The only remaining condition to worry about is that the opamp outputs don't go too high when the dac chip is powered off. With that in mind I put a lot of filter capacitance before the regulators for the 1.8/3.3/5 volt rails shared by all dac channels (didn't do anything for rails unique to each channel). Don't know if the opamps ever glitch high though, and I haven't had the nerve to test what would happen if they did.
Reason I worry about it is I have the I/V opamps separately powered at +-15v with their own non-LDO regulators. Sounded better that way. 7805 regulators sounded better than LDOs for VDD too. Just sayin' - YMMV.
There is some more at the bottom of page 88 which talks about possible damage to the chip and or muting of pops at power up/down.
I thought the I/V feedback routing might have to do with above sorts of concerns, but they never directly say so in the data sheet.
They did include reverse Schottky diodes on the eval board to prevent the opamp outputs from going below ground (on the filtered side of VREF).
The only remaining condition to worry about is that the opamp outputs don't go too high when the dac chip is powered off. With that in mind I put a lot of filter capacitance before the regulators for the 1.8/3.3/5 volt rails shared by all dac channels (didn't do anything for rails unique to each channel). Don't know if the opamps ever glitch high though, and I haven't had the nerve to test what would happen if they did.
Reason I worry about it is I have the I/V opamps separately powered at +-15v with their own non-LDO regulators. Sounded better that way. 7805 regulators sounded better than LDOs for VDD too. Just sayin' - YMMV.
Last edited:
Looking at the internal diagram on the datasheet - I dont see how the internal FB connection helps... it will not solve the power on thump - if the datasheet is to be believed the opamp is effectively unity gain until the AK4499 powers up and enables the Array.
I'm not sure why AKM designed the array like this - atleast with the ESS its Array is never in a HiZ state..
Protection diodes are critical especially if outputs are to be directly paralleled as the I/V stage can source significant current - and you don't want this current to flow though the AK4499 ESD protection circuit on its output pins.
Last edited: