100uf cap looks like one way to get rid of the DC offset at the output of the IV opamps. If those caps are leaky then that could leave some average offset. If so they could be replaced with some decent bipolar audio electrolytics.
There are probably a number of things that could improved. For one quick example, the standard output stage opamp for these dacs is OPA1612.
There are probably a number of things that could improved. For one quick example, the standard output stage opamp for these dacs is OPA1612.
Last edited:
Opamps are already in the works.
I have some OPA1612s I can take out of a dying SMSL M8 (OLED is super shot), and my power amps each have an OPA1656 for the SE input (unused / removed from circuit) which I can harvest. I'm thinking the 1656 will comprise the I/V stage since that seems to be what it is built for and the 1612s can go in the follower. Or do I have that reversed?
I gave a friend a couple LME49990s to use in a headamp he no longer uses. Maybe I can get those back if they would be any better.
Any suggestions on the caps or the opamps? The PCBs on order have both SOIC and SSOP pads. I am going to remove the DIP sockets and couple via pin headers which will be soldered to the boards. I'm not huge on opamp rolling and desoldering headers is easy.
I have some OPA1612s I can take out of a dying SMSL M8 (OLED is super shot), and my power amps each have an OPA1656 for the SE input (unused / removed from circuit) which I can harvest. I'm thinking the 1656 will comprise the I/V stage since that seems to be what it is built for and the 1612s can go in the follower. Or do I have that reversed?
I gave a friend a couple LME49990s to use in a headamp he no longer uses. Maybe I can get those back if they would be any better.
Any suggestions on the caps or the opamps? The PCBs on order have both SOIC and SSOP pads. I am going to remove the DIP sockets and couple via pin headers which will be soldered to the boards. I'm not huge on opamp rolling and desoldering headers is easy.
I would suggest to use OPA1612 for all output stage opamps. Why? Because they seem to tolerate the RF mixed in with audio coming coming out of modern oversampling dac outputs. RF demodulation and remodulation with audio signal effects are not necessarily clearly spelled out in datasheets. However ESS and AKM use OPA1612 for their highest performance dac chips. So do most commercial dac manufacturers. IME if you can't make one of these dacs sound world class with OPA1612 then the problem probably isn't with the opamp.
Also, because of the RF mixed in with the audio output, it is better to get rid of sockets and keep lead inductances low. Using 805 SMD film caps and or NPO/C0G (minimum 50v rated for ceramics for better linearity) bypass caps may help SQ, but if you would like to try that it would probably be wise to check for bypass cap ringing. Its requires a low capacitance active scope probe and a fairly high bandwidth scope to do properly.
Other areas that may be problematic include having suitable-for-the-load voltage regulators dedicated to each dac chip rail, and to each clock, MCU, and opamp rail. Good quality passive components can make a difference too. Same for proper layout with minimum 4-layer PCBs. Pretty much everything matters a whole lot to get excellent performance/SQ out of modern 24-bit data converters. Just a good dac chip and some opamp swapping tends not to get very far.
Also, because of the RF mixed in with the audio output, it is better to get rid of sockets and keep lead inductances low. Using 805 SMD film caps and or NPO/C0G (minimum 50v rated for ceramics for better linearity) bypass caps may help SQ, but if you would like to try that it would probably be wise to check for bypass cap ringing. Its requires a low capacitance active scope probe and a fairly high bandwidth scope to do properly.
Other areas that may be problematic include having suitable-for-the-load voltage regulators dedicated to each dac chip rail, and to each clock, MCU, and opamp rail. Good quality passive components can make a difference too. Same for proper layout with minimum 4-layer PCBs. Pretty much everything matters a whole lot to get excellent performance/SQ out of modern 24-bit data converters. Just a good dac chip and some opamp swapping tends not to get very far.
Yeah I'm not trying to make this SOTA or anything even close to what SOTA (for the 9038Q2M) would even be. I'd be willing to put maybe.... 20-30USD into it (hence scavenging opamps from places they exist but are not used). If I was seriously worried about SQ I'd have just skipped this and put in my Gustard X20 which has been sitting in the closet since 2019. But I'll be real. I sit below a projector when using this sound system so.... yeah. Different setup for critical listening.
That said, both my X20 and DX7 Pro surpass my personal sonic capabilities. The real competitor to this small cheap DAC is my Marantz NR1711 (with the PCM5102A instead of the earlier AKM 55XX series DAC) so not an especially high bar. The closet Gustard is just so stinkin' big. I just wanted to make sure I wasn't missing something that could damage my downstream devices while also not upsetting my 110dB/Wm horns.
So just to be certain I follow, you are saying drop the 100uF electrolytics and replace with NP0/C0G of 50V+ rating? Though I do not have any active/differential probes and my scope is limited to 150MHz. I was actually looking at some Al/Poly caps. Though the ceramics are appealing because they're easily found under 10% tolerance.
My other thought which might (?) go a long way is swapping the Richtek LDOs for AVCC to something like a TPS7A20. < .02uV/(Hz)^1/2
versus the Richtek's 100-200uV noise level. Having an LDO for each DAC chip is a nice little touch, but the noise of the one they chose is...... about 20-40 times higher.
Actually, I just had an idea.... let me see if I can decode the SMSL M8 LDOs. I know it measured quite well for the price. Probably helped having 3x OPA1612s in the line.
That said, both my X20 and DX7 Pro surpass my personal sonic capabilities. The real competitor to this small cheap DAC is my Marantz NR1711 (with the PCM5102A instead of the earlier AKM 55XX series DAC) so not an especially high bar. The closet Gustard is just so stinkin' big. I just wanted to make sure I wasn't missing something that could damage my downstream devices while also not upsetting my 110dB/Wm horns.
So just to be certain I follow, you are saying drop the 100uF electrolytics and replace with NP0/C0G of 50V+ rating? Though I do not have any active/differential probes and my scope is limited to 150MHz. I was actually looking at some Al/Poly caps. Though the ceramics are appealing because they're easily found under 10% tolerance.
My other thought which might (?) go a long way is swapping the Richtek LDOs for AVCC to something like a TPS7A20. < .02uV/(Hz)^1/2
versus the Richtek's 100-200uV noise level. Having an LDO for each DAC chip is a nice little touch, but the noise of the one they chose is...... about 20-40 times higher.
Actually, I just had an idea.... let me see if I can decode the SMSL M8 LDOs. I know it measured quite well for the price. Probably helped having 3x OPA1612s in the line.
Okay. In that case it would probably make sense just to start with the DC offset. First thing might be to verify it is real, and the same for both channels. A DVM might be more accurate than a scope for that. If there is an offset and removing one cap fixes the offset on that channel, then probably leaky caps are the issue.
If the caps are a problem they could be replaced with something else. If you use an electrolytic it will have a sound and it may take a week of running for the sound to settle into whatever its going to be. So first thing when trying caps is don't listen to it for a few days because whether or not you like it at that point shouldn't be what you use to decide if its okay.
One possibility might be Nichicon ES Muse Bipolar audio caps, something like: https://www.amazon.com/pcs-Nichicon-ES-MUSE-Capacitors/dp/B075K4BYF4 ... they should probably be a fairly large value similar to what's in there now. Reason is because if there a DC blocking cap, and then a non-inverting buffer, then there has to be a resistor to ground after the cap and from the opamp noninverting input to ground in order to have a DC path for the opamp input bias current. That resistor and the cap form a high pass filter hopefully with a fairly low corner frequency. For home theater or gaming where a lot of bass might be wanted then the pole might be placed down aaround 5Hz or 10Hz, since it will also affect bass above that frequency to some extent. Another cap that might be good for that would be Panasonic FM 100uf. Again they take about a week of running to settle into how they will sound. They can be bypassed with a .01uf polypropylene Wima film cap if you think the highs are a little weak. If you do try a polarized electrolytic, the + end should go to the I/V opamp because that normally has a small + DC offset.
If you change the caps and there is still an offset, then we would have to look at what's causing it. Could be opamp noninverting input resistor is to big or that it needs to be balanced with a equal value resistor in series with the inverting input in order to deal with opamp input bias current resistor voltage drops.
Once the DC offset thing is resolved, then you can see how you like the sound and see if more work is likely to be helpful.
Regarding the bypass caps I mentioned in the last post they would be for +-15v power rail bypass at the opamp power pins. Usually X7R ceramic is used for that, but if you like the sound with the caps as they are then probably no need to start chipping away at every little thing that might help improve the sound a little. It might take a lot of little fixes to make a very big difference.
You may end up finding you like the old Marantz sound better anyway.
EDIT: Regarding TPS7A20, its can be a reasonably good sounding LDO (as LDOs go). Not all of them sound as good, and AVCC is very sensitive to regulator sound, not just its noise. So if considering a regulator upgrade for AVCC, that might be one to try. Also, with LDOs sometimes sound can be improved at the expense of dumping some power. The trick is to connect a resistor from the LDO output to ground to get some more current flowing through the LDO. That might improve regulator loop gain. For 3.3v AVCC, maybe try 47R, 1/2 watt to ground. Or if that overheats the LDO or makes it go into self-protect mode, maybe something a bit higher like 68R or 100R. Might have to experiment to see if it helps and if so what's a good value.
If the caps are a problem they could be replaced with something else. If you use an electrolytic it will have a sound and it may take a week of running for the sound to settle into whatever its going to be. So first thing when trying caps is don't listen to it for a few days because whether or not you like it at that point shouldn't be what you use to decide if its okay.
One possibility might be Nichicon ES Muse Bipolar audio caps, something like: https://www.amazon.com/pcs-Nichicon-ES-MUSE-Capacitors/dp/B075K4BYF4 ... they should probably be a fairly large value similar to what's in there now. Reason is because if there a DC blocking cap, and then a non-inverting buffer, then there has to be a resistor to ground after the cap and from the opamp noninverting input to ground in order to have a DC path for the opamp input bias current. That resistor and the cap form a high pass filter hopefully with a fairly low corner frequency. For home theater or gaming where a lot of bass might be wanted then the pole might be placed down aaround 5Hz or 10Hz, since it will also affect bass above that frequency to some extent. Another cap that might be good for that would be Panasonic FM 100uf. Again they take about a week of running to settle into how they will sound. They can be bypassed with a .01uf polypropylene Wima film cap if you think the highs are a little weak. If you do try a polarized electrolytic, the + end should go to the I/V opamp because that normally has a small + DC offset.
If you change the caps and there is still an offset, then we would have to look at what's causing it. Could be opamp noninverting input resistor is to big or that it needs to be balanced with a equal value resistor in series with the inverting input in order to deal with opamp input bias current resistor voltage drops.
Once the DC offset thing is resolved, then you can see how you like the sound and see if more work is likely to be helpful.
Regarding the bypass caps I mentioned in the last post they would be for +-15v power rail bypass at the opamp power pins. Usually X7R ceramic is used for that, but if you like the sound with the caps as they are then probably no need to start chipping away at every little thing that might help improve the sound a little. It might take a lot of little fixes to make a very big difference.
You may end up finding you like the old Marantz sound better anyway.
EDIT: Regarding TPS7A20, its can be a reasonably good sounding LDO (as LDOs go). Not all of them sound as good, and AVCC is very sensitive to regulator sound, not just its noise. So if considering a regulator upgrade for AVCC, that might be one to try. Also, with LDOs sometimes sound can be improved at the expense of dumping some power. The trick is to connect a resistor from the LDO output to ground to get some more current flowing through the LDO. That might improve regulator loop gain. For 3.3v AVCC, maybe try 47R, 1/2 watt to ground. Or if that overheats the LDO or makes it go into self-protect mode, maybe something a bit higher like 68R or 100R. Might have to experiment to see if it helps and if so what's a good value.
Last edited:
well..... I measured some things.
The LDOs have on average a 100uV Peak to peak jitter. ESS specs AVCC as +/-5% from 3.3v. I am wondering, since the DAC takes AVCC, and since part of AVCC also goes to the non-inverting opamp input, it seems that the differential nature of the opamp
The XLR outputs, without a source, have on average a 10-12mV noise. Considering the >50k input impedance of the next device, I am not worried. It's not even DC. Just noise. Probably from the relay coils right next to the xlr pins? This time I used chassis ground as GND ref.
Swapping the 4580 out of the followers (5532s it came with were fake or insanely fragile, one pin bent on one and killed the opamp, or DOA.) and replacing with 5532D showed it drop a little. Tested both at full volume and at 0 volume.
Huh. Weird how that offset shows on the scope but not when taking a few things out of the picture. I probably just needed to practice better ground point selection.
Measured the STM branded LDOs on the SMSL. About the same level of noise as this guy. I'll take your cap advice into consideration and decide if I really want to try hand soldering some 10 cent LDOs. I am 90% sure these 'gold" caps are fakes, and probably really poor ones as well.
Since the I/V stage is set so hot now, I am way less worried about single digit mV ripple. No way that will be audible above thinks like real life.
TI-TINA even shows the opamps unfazed by 5mv ripple on the non-inverting input.
Again, thank you very much for the help and explaining what should be where. That makes interpreting this circuit and layout 1000x easier since I am not a DAC designer nor an EE,
ETA: These speakers have NEVER struggled with highs.
The LDOs have on average a 100uV Peak to peak jitter. ESS specs AVCC as +/-5% from 3.3v. I am wondering, since the DAC takes AVCC, and since part of AVCC also goes to the non-inverting opamp input, it seems that the differential nature of the opamp
The XLR outputs, without a source, have on average a 10-12mV noise. Considering the >50k input impedance of the next device, I am not worried. It's not even DC. Just noise. Probably from the relay coils right next to the xlr pins? This time I used chassis ground as GND ref.
Swapping the 4580 out of the followers (5532s it came with were fake or insanely fragile, one pin bent on one and killed the opamp, or DOA.) and replacing with 5532D showed it drop a little. Tested both at full volume and at 0 volume.
Huh. Weird how that offset shows on the scope but not when taking a few things out of the picture. I probably just needed to practice better ground point selection.
Measured the STM branded LDOs on the SMSL. About the same level of noise as this guy. I'll take your cap advice into consideration and decide if I really want to try hand soldering some 10 cent LDOs. I am 90% sure these 'gold" caps are fakes, and probably really poor ones as well.
Since the I/V stage is set so hot now, I am way less worried about single digit mV ripple. No way that will be audible above thinks like real life.
TI-TINA even shows the opamps unfazed by 5mv ripple on the non-inverting input.
Again, thank you very much for the help and explaining what should be where. That makes interpreting this circuit and layout 1000x easier since I am not a DAC designer nor an EE,
ETA: These speakers have NEVER struggled with highs.
Wondering how you are measuring noise? With a scope?
Also, a possible thought about the average voltage shown on the scope readouts. Could be the I/V opamp output swing is enough to reverse bias the 100uf electrolytic coupling caps on negative peaks. In that case the waveform might be sort of compressed on one side (low order even harmonic distortion), which can be hard to see on a scope just from eyeballing the waveform. If so, probably less average offset on the scope if the volume level is turned down to where the I/V opamp output swing is limited to +-1.5v
In that case switching to the Nichicon bipolar electrolytics might reduce the distortion on louder peaks. However, the distortion may or may not sound bad to you. In fact that dac may overall sound better with the distortion there if it is helping to mask other lower level distortions.
In that regard, sometimes fixing dac problems can be kind of like peeling away layers of an onion. Fix one problem only to better expose the next lower level problem 🙂
Also, a possible thought about the average voltage shown on the scope readouts. Could be the I/V opamp output swing is enough to reverse bias the 100uf electrolytic coupling caps on negative peaks. In that case the waveform might be sort of compressed on one side (low order even harmonic distortion), which can be hard to see on a scope just from eyeballing the waveform. If so, probably less average offset on the scope if the volume level is turned down to where the I/V opamp output swing is limited to +-1.5v
In that case switching to the Nichicon bipolar electrolytics might reduce the distortion on louder peaks. However, the distortion may or may not sound bad to you. In fact that dac may overall sound better with the distortion there if it is helping to mask other lower level distortions.
In that regard, sometimes fixing dac problems can be kind of like peeling away layers of an onion. Fix one problem only to better expose the next lower level problem 🙂
On the Marantz... nah, been there done that. I got this receiver because my old Sony transformer started to de-lam and 4k60+ is finally hitting mainstream. I used to be happy with the Sony until I got a DX7 Pro and a pair of JBL 308p MkII's for my PC setup. Even then, the Sony was just a cheap HDMI switch, big amp, and volume pot (using the SMSL M8). The Marantz was a factory refurb. If I were to pay full price, I'd look at the Tonewinner or Emotiva AVPs.
No room for anything even nearing 1/2W in power by the LDOs unless it were to overhang LVDS lines, opamps, crystal, or the ESS chips. Those AVCC LDOs are SOT23-5. I am guessing by 'sound' you mean the transient response and VDroop vs current load and thermals combined with Vresponse (PSRR) on upstream? My probes when floating have about 80mV they pick up in that room due to a dual-band AP on the other side of the wall, multi-ganged z-wave switches, and lots of electronics in their own connected sleep mode. I can increase the BG noise by making a loop with the lead. So I mentally note the noise and subtract it.
AH, actually I think we can rule out the compression by using the measurements from the DSO:
Vtop: 4.6V
Vbot: -3.8V
Vmid: .4V
Vtop-Vmid = 4.2V. Vmid - Vbot = -4.2V
Good to hear of noise subtraction. Would just check with you that you know that its good scope practice to the ground probe on the closest ground to where the probe tip will go, and that once a probe ground point is chosen, touch the probe tip to the ground clip connected at that ground point to get the common mode ground noise as it affects the scope at that ground point. Its all common knowledge anyway (starting around page 46 IIRC): https://download.tek.com/document/02_ABCs-of-Probes-Primer.pdf ...only thing I noticed missing in the latest version of that document is nothing said about spring grounds. Don't know why they left that out.
Regarding compression, what the scope readings say doesn't seem incompatible with that I suggested might be the cause (asymmetrical compression of peak voltages as viewed after DC removal by a polar electrolytic cap, given the particular AC and DC signal levels -- we could talk about it more if anyone cares). Maybe some of the other guys who follow the thread will offer some other possibilities. Then again, up to you if you care and or have interest in poking around a little more or not. Perfectly fine whatever you decide.
Last edited:
My cheap scope reinforces this with a ground clip that can barely reach the measuring probe.Good to hear of noise subtraction. Would just check with you that you know that its good scope practice to the ground probe on the closest ground to where the probe tip will go, and that once a probe ground point is chosen, touch the probe tip to the ground clip connected at that ground point to get the common mode ground noise as it affects the scope at that ground point. Its all common knowledge anyway (starting around page 46 IIRC): https://download.tek.com/document/02_ABCs-of-Probes-Primer.pdf ...only thing I noticed missing in the latest version of that document is nothing said about spring grounds. Don't know why they left that out.
Regarding compression, what the scope readings say doesn't seem incompatible with that I suggested might be the cause (asymmetrical compression of peak voltages as viewed after DC removal by a polar electrolytic cap, given the particular AC and DC signal levels -- we could talk about it more if anyone cares). Maybe some of the other guys who follow the thread will offer some other possibilities. Then again, up to you if you care and or have interest in poking around a little more or not. Perfectly fine whatever you decide.
On the italics, oh snap I did not know that and it makes a ton of sense. I do think I habitually do that just to watch the scope's own noise drop and make sure the gound is sinking properly.
I'm always interested in poking around and learning, though I am constantly reminded of my lack of formal training (not an EE), and always humbled further when I try and solder.
I got the OPA1612's in the IV stage today. Dropped the DIP sockets. Removed the HP amp receptacles (they are convenient test pads) and I swapped the cheap 'gold' Elna caps for Oscon CS polymers. Topping and SMSL put these caps everywhere in the signal path (and some tantalums which wouldnt fit) and those DACs measure exceptionally well so I figure they can;t be any worse than the questionable Elna caps. Not sure if I will leave the NE5532s or if I will try the 1656s there.
Maybe I will swap those richtek LDOs. 70nV less noise on the AVCC ckt seems like it might be a good swap, but I am unsure how much those will really impact the chips in current mode. Luckily my donor SMSL M8 board has a few ST LD3985s I can steam to swap the Richtek RT9193s.
Removing the plastic caps from the bottom on the oscons was really fiddly. Maybe not BP, but that was the only way they would fit.
I am hoping/guessing they will fall somewhere between film and biased electrolytic on figure 4. https://www.ti.com/lit/an/slyt796/s...04108&ref_url=https%3A%2F%2Fwww.google.com%2F
Well, oopsie daisy. That'll teach me to burn the midnight solder. Right channel XLR pin 2 outputting a 5-20Vpp square wave xlr pin 3 putting out an 8Vpp sine. Left channel Pins 2&3 4.4Vpp centered at 0V.
Oh and I figured out the "offset" using the method you described. Ground pin on XLR pin 1 and probe tip to ground clip = 240mV reading. Examining the XLR connectors shows that the locking tab is floating. It never makes it to circuit ground nor the enclosure. Stupid design, considering most commercial cables will bridge pin 1 and connector housing..
First suspect on the issues: Caps. The polymers had fragile legs and not every leg made it through the board. I was hoping maybe a loose connection. I did find what I was hoping for, but it was not the root cause. Since the only 6.3mm branded caps I had on hand were Nichicok PS series, I went with those. They are used for low leakage SMPS duty normally (per Nichicon's datasheet) but as just a dc-decoupler they seemed fine for now. I also found some .047uF film caps and parallelled those on the bottom in case the electrolytics awere not full up to the job. Since they are 25V parts (electros) and 63V films of unknown origin, I just wanted the dang thing to work. Shown here before I cleaned up the solder braid flux carefully with q-tips and a tungsten scribe (used flat bevel on the conical tip to lift big flux crust)
Removing the polycaps showed the true quality of this board..... or complete lack thereof. The pads are falling off at 350C under a few seconds of heat. I put male pins through the holes and scraped away some mask (alarmingly easy) to re-create the connections.
Tracing back the bad weaveform from the outputs, only the bottom (in photo) OPA1612 had a square out on one pin (Pin #1) and the abnormally high (11Vpp) sine wave on the other. The top Opamp pair were giving identical outputs (to each other, but lower than design would suggest is correct) so I thought maybe I damaged it (bottom OPA1612) with heat. Swapped it out for a fresh OPA1612 and..... same.
Tracking and comparing each channel's pin, found pin 6 on the fussy OPA1612 had a really awful looking wave. Not smooth, not square, etc. The LDO near the bottom 9038 also showed an output near 4V and the top LDO showed an output near 2V. This was strange, and these should both be 3.3V units. Pin 6 connects direct to the ES9038q2M, but the DACR and DACR_B are fed via AVCC so I thought hey, maybe the LDOs are somehow doing a thing.
So top LDO appeared to be a problem (under 2V output) and I swapped that with the single 3.3V STM LDO I had been able to salvage. HEre's where things got stranger. This brought the lower opamp completely into line. Exactly 8.4Vpp on both outputs. The virtual grounds of both are connected through one cap per LDO. The Top LDO which shows 0V on all pins is connected to the 5V line for the relays. The relays obviously work, because I can get waveform from them. All three are connected in parallel.
I think I kept it the right way up, but it had been sitting under a pool of semi-hardened flux for a decade so the markings are of course completely gone. This resulted in perfect 8.4VPP behavior on the bottom opamp set and....a completely dead top (Left) channel.
I admit, this one has me stumped. There is continuity between the relay coils and the v_in of the top LDO, but I am seeing 0V on said LDO at the EN/V_IN pin yet the relays can be heard latching.
Oh and I figured out the "offset" using the method you described. Ground pin on XLR pin 1 and probe tip to ground clip = 240mV reading. Examining the XLR connectors shows that the locking tab is floating. It never makes it to circuit ground nor the enclosure. Stupid design, considering most commercial cables will bridge pin 1 and connector housing..
First suspect on the issues: Caps. The polymers had fragile legs and not every leg made it through the board. I was hoping maybe a loose connection. I did find what I was hoping for, but it was not the root cause. Since the only 6.3mm branded caps I had on hand were Nichicok PS series, I went with those. They are used for low leakage SMPS duty normally (per Nichicon's datasheet) but as just a dc-decoupler they seemed fine for now. I also found some .047uF film caps and parallelled those on the bottom in case the electrolytics awere not full up to the job. Since they are 25V parts (electros) and 63V films of unknown origin, I just wanted the dang thing to work. Shown here before I cleaned up the solder braid flux carefully with q-tips and a tungsten scribe (used flat bevel on the conical tip to lift big flux crust)
Removing the polycaps showed the true quality of this board..... or complete lack thereof. The pads are falling off at 350C under a few seconds of heat. I put male pins through the holes and scraped away some mask (alarmingly easy) to re-create the connections.
Tracing back the bad weaveform from the outputs, only the bottom (in photo) OPA1612 had a square out on one pin (Pin #1) and the abnormally high (11Vpp) sine wave on the other. The top Opamp pair were giving identical outputs (to each other, but lower than design would suggest is correct) so I thought maybe I damaged it (bottom OPA1612) with heat. Swapped it out for a fresh OPA1612 and..... same.
So top LDO appeared to be a problem (under 2V output) and I swapped that with the single 3.3V STM LDO I had been able to salvage. HEre's where things got stranger. This brought the lower opamp completely into line. Exactly 8.4Vpp on both outputs. The virtual grounds of both are connected through one cap per LDO. The Top LDO which shows 0V on all pins is connected to the 5V line for the relays. The relays obviously work, because I can get waveform from them. All three are connected in parallel.
I think I kept it the right way up, but it had been sitting under a pool of semi-hardened flux for a decade so the markings are of course completely gone. This resulted in perfect 8.4VPP behavior on the bottom opamp set and....a completely dead top (Left) channel.
I admit, this one has me stumped. There is continuity between the relay coils and the v_in of the top LDO, but I am seeing 0V on said LDO at the EN/V_IN pin yet the relays can be heard latching.
The pin you have labeled EN in the above pic looks like its where power comes in. C24 and L3 appear to be the input power filter. C27 is probably the LDO output cap. Is the red curved line to indicate pins are connected together under the chip? Also, looks like there should be +5v on both sides of L3, as it is series the LDO input power.
Its important to sketch up a schematic if possible, and try to identify parts from the case markings before replacing anything. Can't tell but what looks like the LDO pin connected at C25(?) might be a noise filter pin?
There are a few websites that have reverse part number lookup tables that can help identify SMD device part numbers from case markings. Don't know if you still have the original LDOs to see how they're marked.
EDIT: Regarding the opamp waveforms, you might try measuring the voltage at every pin with the scope. May give some clue why the output looks the way it does.
Its important to sketch up a schematic if possible, and try to identify parts from the case markings before replacing anything. Can't tell but what looks like the LDO pin connected at C25(?) might be a noise filter pin?
There are a few websites that have reverse part number lookup tables that can help identify SMD device part numbers from case markings. Don't know if you still have the original LDOs to see how they're marked.
EDIT: Regarding the opamp waveforms, you might try measuring the voltage at every pin with the scope. May give some clue why the output looks the way it does.
Last edited:
This was also my initial thought, but prior to the swap, I compared datasheets:
Ah, found the issue. V_In lead on the salvaged LDO was short/bent compared to others. This lead to a cold joint. I had to use macro and really zoom in. Wow the probe touches look super magnified. Lots of continuity testing 🙄
The Voltage issue was a partially delaminated/cracked ferrite bead upstream of the LDO.
This high R caused the 5.4V rail feeding the top LDO to drop to ~2.XV which Ohm's law shows the LDO is only passing about 7mW (and boosting to about 2.6V). I had some Murata beads of a sliiiightly larger size and swapped those.
SMPS is 56.3KHz (seems pretty common in SMPS) and the Richtek only has about 45dB PSRR there. TI has about 20-30dB advantage in PSRR and about..... 18 times less noise. Richtek datasheet shows the 3V model having ~100-200uV noise output. TI shows a 20mA load having ~7uV noise.
So now everything works as expected but one side (Richtek) is markedly noisier than the other and it shows as weebly wobblies on the peaks. Pending delivery of the TI LDOs and some 49720s for the buffers just because they are actually obtainable. Removed the 5532 on the RCA line because it was just sitting there generating heat for nothing.
Ah, found the issue. V_In lead on the salvaged LDO was short/bent compared to others. This lead to a cold joint. I had to use macro and really zoom in. Wow the probe touches look super magnified. Lots of continuity testing 🙄
The Voltage issue was a partially delaminated/cracked ferrite bead upstream of the LDO.
This high R caused the 5.4V rail feeding the top LDO to drop to ~2.XV which Ohm's law shows the LDO is only passing about 7mW (and boosting to about 2.6V). I had some Murata beads of a sliiiightly larger size and swapped those.
SMPS is 56.3KHz (seems pretty common in SMPS) and the Richtek only has about 45dB PSRR there. TI has about 20-30dB advantage in PSRR and about..... 18 times less noise. Richtek datasheet shows the 3V model having ~100-200uV noise output. TI shows a 20mA load having ~7uV noise.
So now everything works as expected but one side (Richtek) is markedly noisier than the other and it shows as weebly wobblies on the peaks. Pending delivery of the TI LDOs and some 49720s for the buffers just because they are actually obtainable. Removed the 5532 on the RCA line because it was just sitting there generating heat for nothing.
IIRC, LME49720 is the same as LM4562. Just a different part number system they set up for LME opamps. The problem with using those in DACs is that DAC chip analog outputs have a lot of RF in them that has to be filtered out in the output stage. Unfortunately 49720/4562, again IIRC, are sensitive to RF, which can then cause distortion. The opamp of choice for DAC output stages remains OPA1612. Again, if that one doesn't sound good then there are other problems.
One problem might be use of an SMPS. Substituting a linear supply might clean up the sound quite a bit. Only way to find out is try it.
One problem might be use of an SMPS. Substituting a linear supply might clean up the sound quite a bit. Only way to find out is try it.
Ah yeah, not disagreeing on the 1612 > 49720, just well....
I feel okay about the 49720 for now because it is also used in the Topping DX7 Pro which measures exceptionally well, and it feeds my office active speakers and lives beneath a 49" ultrawide screen and about 2 foot from a very very power hungry PC with a wireless AP less than 10 feet away on the ceiling. Also uses an ESS chip. I always have some backup 5532s and 4850s and can steal those 1656's if required.
Regarding PSU... well..... I traced the 5V lines. There is literally no way a better PSU could hurt.
"Power conditioning? No, no, no; conditioner is just a hair product. 5V comes in, the relay does a thing using the 5V to connect 5V and then it becomes 3.3V and goes into the DACs. Magic. What more do you want?"
🤣
Leaf Audio's successor product to this board switched to cirrus logic chips and comes with a 5V linear supply. Maybe somebody did learn.
Topping used LME49720 in D90 too, but not for I/V. Most of the RF filtering was done before the audio got the LME's. Also they were configured for differential summing, but without MFB filtering. In that situation they are probably fine. However, most dac output stages don't look like what Topping did with D90. The standard 3-opamp output stage is shown in many dac chip datasheets. In that circuit for each channel there are a pair of I/V opamps followed by, an opamp doing balanced MFB filtering and differential summing, both at the same time. In that case there is still a lot of RF going into both the I/V and the summing/filtering stages. For that topology IME its better to use OPA1612 for all output stage opamps.
IME LM4562 works very well as I/V op amp for ES9038Q2M. Even better than OPA1612 which seems to have much more chip-to-chip variance. As summing op amp OPA1656 works fine.
AH, I was not clear. 1612s are staying in the I/V in my setup. They are soldered in, and the the math from the datasheet values matches almost perfectly with the measurements of the output at the I/V. This DAC was two opamp stages (3 if you exit via RCA). 49720 will be in the unity gain voltage follower stage for XLR outputs, exactly like my DX7 Pro.
That stage is shown to the right of the OPA1612s in the photo above. The RCA diff summing/conversion stage is offscreen further to the right (barely visible).
TI DOES have an application note mentioning RF filtering on the PCM17XXX DAC with RC filters but I don't think TI also has the SW-configurable LP filter regs that ESS does; which cuts frequencies according to the playback data (PCM vs DSD). I know the ESS filter in DSD mode will reach up to at least 100KHz but unsure if it extends to the MHz range.
In both the I/V and unity buffer stages the risk of RF is also much less from what I can see because Leaf has at least decoupled rails to ground within a couple mm of the pins. I don't think there's any risk of 1/4 wavelength gain . Now whether or not they used a low impedance ground (referenced to the rails) is a whole other matter. 😀
That stage is shown to the right of the OPA1612s in the photo above. The RCA diff summing/conversion stage is offscreen further to the right (barely visible).
TI DOES have an application note mentioning RF filtering on the PCM17XXX DAC with RC filters but I don't think TI also has the SW-configurable LP filter regs that ESS does; which cuts frequencies according to the playback data (PCM vs DSD). I know the ESS filter in DSD mode will reach up to at least 100KHz but unsure if it extends to the MHz range.
In both the I/V and unity buffer stages the risk of RF is also much less from what I can see because Leaf has at least decoupled rails to ground within a couple mm of the pins. I don't think there's any risk of 1/4 wavelength gain . Now whether or not they used a low impedance ground (referenced to the rails) is a whole other matter. 😀
Attachments
Does not worry me at all:
https://www.diyaudio.com/community/...trumentation-applications.347854/post-7021805
https://www.diyaudio.com/community/...trumentation-applications.347854/post-7021805