GreatLaBroski, Thank you for the link. I am sure there are much better instructions for SMD soldering than I can make, but I haven't been able to figure out how to get people over their inertia and start trying some new things and learning more skills.
Of course, it could be some people can't afford a new roll of solder or a new solder iron tip. I don't know if that is the problem or not, people are going to have to tell me if they have problems or not with things like that or whatever it is that keeps them from doing what is suggested, or sometimes very strongly suggested for dac modding. I'm not a mind reader, obviously.
Anyway, not just speaking to you, but to all our interested readers, I need to get some feedback on the things I posted today. What is good and what is bad? What do you need more and what less of?
I know I promised a though-hole output stage, and there are multiple ways I can think of laying one out. The main questions about SMD mostly have to do with whether or not I can find good enough through-hole resistors, and if people would like to use LTC6655 for an AVCC reference.
It would be good if we could keep up a conversation about whether people's confidence can be gotten to the point we could use SMD resistors in critical places if needed for better SQ, and if we could use an SMD IC now and then if it will help SQ. So, I am trying to keep a conversation going on that to see if there is anything I can do to help make progress with what people feel comfortable with, or if we are really going to have to go with super beginner level soldering for everything.
Also, how about using and soldering together adhesive backed copper foil for a ground plane? I think that is necessary for best results if people can possibly manage it. I will need to hear from people to see if they think they can do it.
Of course, it could be some people can't afford a new roll of solder or a new solder iron tip. I don't know if that is the problem or not, people are going to have to tell me if they have problems or not with things like that or whatever it is that keeps them from doing what is suggested, or sometimes very strongly suggested for dac modding. I'm not a mind reader, obviously.
Anyway, not just speaking to you, but to all our interested readers, I need to get some feedback on the things I posted today. What is good and what is bad? What do you need more and what less of?
I know I promised a though-hole output stage, and there are multiple ways I can think of laying one out. The main questions about SMD mostly have to do with whether or not I can find good enough through-hole resistors, and if people would like to use LTC6655 for an AVCC reference.
It would be good if we could keep up a conversation about whether people's confidence can be gotten to the point we could use SMD resistors in critical places if needed for better SQ, and if we could use an SMD IC now and then if it will help SQ. So, I am trying to keep a conversation going on that to see if there is anything I can do to help make progress with what people feel comfortable with, or if we are really going to have to go with super beginner level soldering for everything.
Also, how about using and soldering together adhesive backed copper foil for a ground plane? I think that is necessary for best results if people can possibly manage it. I will need to hear from people to see if they think they can do it.
Last edited:
In other news, oddly to me I am being asked to confirm for the second time in the last week or so that I am still interested in hearing Katana dac version 1.2. Once again, I have responded "Yes!," I am still interested, and I am still engaged actively in my role as a Katana dac reviewer.
Latest email communications with Allo indicate the shipping schedule appears to have slipped from what was originally expected to be this week to sometime next week, but next week should be fine too. I still expect to be right here.
From what I understand at this point there is to be a revised output stage board with the discrete opamp filter sections having lowered corner frequencies. A newly added post dac, but pre-I/V filter is said to be included as well (sometimes referred to as a 'Rasmussen' filter, named after our own diyaudio member Joe Rasmussen).
In addition to all foregoing, an isolator board is expected to be included with the new Katana board which should help keep noise from RPi from getting into the stack of dac boards and possibly adversely affecting SQ slightly, although for some reason it only works up to 192kHz, IIRC. Presumably that limitation has to do with technical issues associated with keeping clock and data signal edges aligned to within spec at higher frequencies. (Misaligned timing of digital signal edges may sometimes be referred to as skew. Hopefully, we will have no skew not-downs this time around
)
Latest email communications with Allo indicate the shipping schedule appears to have slipped from what was originally expected to be this week to sometime next week, but next week should be fine too. I still expect to be right here.
From what I understand at this point there is to be a revised output stage board with the discrete opamp filter sections having lowered corner frequencies. A newly added post dac, but pre-I/V filter is said to be included as well (sometimes referred to as a 'Rasmussen' filter, named after our own diyaudio member Joe Rasmussen).
In addition to all foregoing, an isolator board is expected to be included with the new Katana board which should help keep noise from RPi from getting into the stack of dac boards and possibly adversely affecting SQ slightly, although for some reason it only works up to 192kHz, IIRC. Presumably that limitation has to do with technical issues associated with keeping clock and data signal edges aligned to within spec at higher frequencies. (Misaligned timing of digital signal edges may sometimes be referred to as skew. Hopefully, we will have no skew not-downs this time around
)Hi terry22,
There are a few posts from today discussing possible plans for an output stage and some closer look at SMD soldering possibilities. I also wrote a few things in the ES9038Q2M Production thread that Serge started.
The next thing I want to do is settle on an output stage schematic, something I hope to work on tomorrow. In the meantime I really would like your responses, comments, and questions regarding what I posted today, since you are one of the people who has expressed interest in step-by-step recipes for output stage construction. Are the things I have been writing about helpful, confusing, useless, what? Can you get some small diameter solder and a small solder iron tip?
Thanks in advance for your reply.
----------------------------------------
There are some other people who posted recently that I think may be interested in the same or similar types of things as terry22. I am thinking of the following folks, for example, but not limited to:
Sergelisses
Blackfear
DRONE7
nikosokey
paulmarinis
How am I doing guys? Did I say anything in a useful direction for your interests today? What do you need from me tomorrow? Is okay if I continue to fill in more details of a prospective through-hole output stage? Are you okay with the copper foil ground plane concept? If I continue in this direction will you go ahead and mod? (I know you all didn't ask about same things or post to the thread for the same reasons, but I need to hear from you because I need feedback. If I don't get enough, I will just go off and do whatever and only then you may find out you don't want to build it. That would be too bad, we need to try to get it right the first time.) Thanks!
Last edited:
Looking good Mark ! Through hole works for me as a diy-er but willing to try smd
(as a lifelong jeweller I know all about keeping your solder clean !.......... just not so good at soldering below 1270 degrees F..... lol)
Through hole I/V works for me !
(as a lifelong jeweller I know all about keeping your solder clean !....
...... just not so good at soldering below 1270 degrees F..... lol)Yes! looking forward to it ..!
Through hole I/V works for me !
Last edited:
Copper foil ground plane okay? If yes, can you order some adhesive backed copper foil?
Can you order an SMD solder practice kit? If the answer is no, that's okay, but if willing to try to learn you will need a kit and they are less than $10.
Last edited:
I am waiting for a v1.07 board to arrive in the next weeks so you can count me as another interested party. I am willing to try the OPA1612s in the output stage as they seem to be favoured over the LME49720 in the recent application papers (ordered both of them, just in case...). There are two documents that grabbed my attention when I was browsing the OPA documentation:
HiFi Audio Circuit Design
http://www.ti.com/lit/pdf/tidu672
(please, ignore the headphone amp part and focus on the I/V LPF part)
There are two concepts that I might have missed to see in this, or another for that matter, discussion:
1. Better THD if the transimpedance amplifier outputs signal at maximum voltage swing.
2. If the output voltage of the DAC is less than AVCC then why should we choose 1,65V as the reference voltage?
As for the technical part - I would prefer to start with SMD and switch to THD only if it becomes a total sloppy disaster.
HiFi Audio Circuit Design
http://www.ti.com/lit/pdf/tidu672
(please, ignore the headphone amp part and focus on the I/V LPF part)
There are two concepts that I might have missed to see in this, or another for that matter, discussion:
1. Better THD if the transimpedance amplifier outputs signal at maximum voltage swing.
2. If the output voltage of the DAC is less than AVCC then why should we choose 1,65V as the reference voltage?
As for the technical part - I would prefer to start with SMD and switch to THD only if it becomes a total sloppy disaster.
Hi KoAP, Sure, agree it would be possible to use a battery for an approximately 3.2v reference. Not sure that approach would be my first choice though because batteries require maintenance, and do not supply a constant voltage over time as they discharge. Of course, the good thing is they are not affected by power supply noise. However, something like LTC6655 can work really well and is probably even better than a battery to my way of thinking about all the trade offs.
There is something curious about the schematic in your post though, which is that I am wondering what the opamp with the diode connected to it is doing for the circuit.
longimanus, Yes, I am also well aware of OPA1612. I think the main reason they are being used instead of LME49720 or LME49860 is because they can supply a little bit higher output current which is necessary for use with ES9038PRO. Other than output current, however, I don't think they are superior to LME49720. Also, LME49720 can supply enough current for use with ES9038Q2M. That being said, you could use either type of opamp if you want.
Regarding the other couple of issues you mention, please let me respond to each one. (1) The only reason they say THD is better with higher output swing is because less of the output signal time is spent down near ground where opamp distortion is higher because of their internal opamp cross over distortion. But that distortion is produced at every zero crossing which happens all the time, so it's always there. That is one of the drawbacks with using opamps. But, low average THD per cycle, or per zero crossing, in one stage is not necessarily the best choice in a multistage circuit. For example, the gain structure that results in less noise may be better in some ways than the gain structure than results in the least average THD per cycle of audio output in one single stage. In fact, ESS does test their recommended circuits for lowest overall distortion and noise measured where the signal comes out of the last opamp. To get the best performance there, I believe the I/V stage part of the circuit is better run at a little lower voltage swing.
And, (2) 1.65v is used as an offset because it is about 1/2 of AVCC, which would be the quiescent operating point of the two differential output signals for each dac channel. When there is zero output from the dac, each output from the dac should be offset to about 1.65v (exactly 1/2 AVCC, actually).
In addition, each of the two balanced dac output signals per channel have a lot of common mode noise and distortion on them. We need to remove the common mode signal and only keep the differential signal. For that reason we use two 1.65v offset virtual ground I/V converters to first convert the offset differential current signals to voltages, then take the difference between the two.
At the same time as we are doing the above processing with the dac output currents, we need to finish reconstruction filtering and remove RF leakage that also comes out of the dac outputs. Using opamps that are tolerant of RF is a good start, and LME49720 have a higher open loop bandwidth than OPA1612, likely allowing better behavior at high frequencies above audio which are present along with the audio signals.
In addition, the simplified dac output filtering shown in application note simulations fails to take into account the need to filter out RF above the frequencies where the opamps actually have enough gain to work like we usually expect from opamps. For best sound quality results including as the power amp and speakers may be affected, we should have some passive filtering sections in the output stage in addition to the active filtering given by the opamp sections. In short, empirically we tend to find that more thorough filtering often results in better sound quality coming out of the speakers. For the above reasons, the output stage schematic I will recommend using will include more than the theoretical minimum filtering that would be needed if reconstruction filtering were the only concern.
Regarding the other couple of issues you mention, please let me respond to each one. (1) The only reason they say THD is better with higher output swing is because less of the output signal time is spent down near ground where opamp distortion is higher because of their internal opamp cross over distortion. But that distortion is produced at every zero crossing which happens all the time, so it's always there. That is one of the drawbacks with using opamps. But, low average THD per cycle, or per zero crossing, in one stage is not necessarily the best choice in a multistage circuit. For example, the gain structure that results in less noise may be better in some ways than the gain structure than results in the least average THD per cycle of audio output in one single stage. In fact, ESS does test their recommended circuits for lowest overall distortion and noise measured where the signal comes out of the last opamp. To get the best performance there, I believe the I/V stage part of the circuit is better run at a little lower voltage swing.
And, (2) 1.65v is used as an offset because it is about 1/2 of AVCC, which would be the quiescent operating point of the two differential output signals for each dac channel. When there is zero output from the dac, each output from the dac should be offset to about 1.65v (exactly 1/2 AVCC, actually).
In addition, each of the two balanced dac output signals per channel have a lot of common mode noise and distortion on them. We need to remove the common mode signal and only keep the differential signal. For that reason we use two 1.65v offset virtual ground I/V converters to first convert the offset differential current signals to voltages, then take the difference between the two.
At the same time as we are doing the above processing with the dac output currents, we need to finish reconstruction filtering and remove RF leakage that also comes out of the dac outputs. Using opamps that are tolerant of RF is a good start, and LME49720 have a higher open loop bandwidth than OPA1612, likely allowing better behavior at high frequencies above audio which are present along with the audio signals.
In addition, the simplified dac output filtering shown in application note simulations fails to take into account the need to filter out RF above the frequencies where the opamps actually have enough gain to work like we usually expect from opamps. For best sound quality results including as the power amp and speakers may be affected, we should have some passive filtering sections in the output stage in addition to the active filtering given by the opamp sections. In short, empirically we tend to find that more thorough filtering often results in better sound quality coming out of the speakers. For the above reasons, the output stage schematic I will recommend using will include more than the theoretical minimum filtering that would be needed if reconstruction filtering were the only concern.
Last edited:
Referring back to here: http://www.diyaudio.com/forums/digital-line-level/314935-es9038q2m-board-247.html#post5533102
Okay, I am a little more awake now and it looks like the opamp with diode is to respond quickly to sudden increased current draw from the AVCC circuit loads. So far as I am aware that is not problem we need to do anything about, but we have not fully characterized the behavior of AVCC as a load at HF. If we did some careful measurements and decided we needed to speed up AVCC power supply loop regulation at HF then we might consider doing something maybe a little bit fancy and unusual to fix the problem. So far though, I am not aware that such a need exists or that sound quality could be helped in any way by going to such efforts, so I think I will keep my AVCC circuits about the same as I usually do for now. Also, not sure why AVCC current draw transients should cause a compensation response in DVDD or Clock supply voltages, which doesn't quite make sense. Thank you for sharing the design idea though. Will keep something like in mind in case I do need it someday.
Last edited:
There is something curious about the schematic in your post though, which is that I am wondering what the opamp with the diode connected to it is doing for the circuit.[/QUOTE]
Markw4, if you compare the amplitude- freqency response of the circuits in the post #1146 and #2468, it will be clear for what opamp with the diode.
Markw4, if you compare the amplitude- freqency response of the circuits in the post #1146 and #2468, it will be clear for what opamp with the diode.
I guess that this is what might happen at the end - two identical circuits with both types of op amps.. if I have enough time
Now there's the thing. The Vout range of the DAC = k.AVCC, where k is a little < 1 (value stated in the datasheet, don't know if I'm allowed to cite it publicly, although it is widely available), or somewhere around 3V. Let's assume that the zero out is AVCC/2 (this value has to be generated inside the DAC, before the output, right?) then the maximum + swing should be VoutP-P - AVCC/2 = k.AVCC - AVCC/2 ~ 1,35V. That would make the usable P-P output ~ 2,7V and leave a dead space of ~ 300mV. That makes the next question more important. (Please, tell me where I've got my calculations wrong)
If the usable signal therefore reaches at best an amplitude of 2,7V then we could probably use a little more amplification than planned for 3,3V? I can see that in the ESS 'every budget' paper the feedback network is with comparable resistance to what is suggested in the OPA16xx evaluation schematic (2k+), even with higher Zout of the ES9008.
Concerning high frequency filtering: I have seen mentions on the required use of LPF with corner frequency ~50kHz in systems reproducing DSD, and most of the LPFs I have seen here and in the application notes of different opamps start at several hundred kHz. Maybe this one also needs a little thought.
Last edited:
Sorry, I just deleted the last post because I got into a situation of mixing together two different parts of the output stage. But to respond to longimanus, it is possible to run AVCC at a little higher than 3.3v and perhaps derive some benefit from that, although ESS seems to be strongly recommending 3.3v. It may be they are doing some internal optimization for it they don't talk about, don't know.
Anyway, output stages are typically comprised of subsections such as I/V and differential. They each bring their own issues. The I/V opamp runs with some common mode input offset, and typically is designed with a few small volts of output swing maybe 3v peak to peak. They also typically run into low impedances which might be too low for some output swings. Low impedances can help keep noise low, and low output swings can help keep the I/V opamps from getting into cross over where distortion rises. Graphs of opamp output distortion seem to be taken by manufacturers with fixed test signals and in part reflect the amount of time a test sine wave will spend in the cross over region. There are also graphs of distortion vs output load resistance which show low load resistance causes distortion. Both the graphs may be of limited applicability for opamps with limited output swings with output offset to avoid the cross over region.
While it may be possible to increase AVCC closer to maximum limits and gain a little benefit, it is not likely that is where most output stages produce most of their distortion.
I/V stages are typically followed by differential summing circuits that operate over greater common mode input ranges than the I/V stages, and they also remove DC offsets from the I/V stages so that the final dac output can be near ground level. Having the output near ground level can allow removal of output capacitors and or output transformers that would otherwise be required.
It could be that most first order analysis of output stage distortion occurs in the differential stage due to the less ideal input common mode offsets and output cross over. An additional source of apparent distortion in output stages appears to come from very low level RF leakage through the dac outputs. If amplified sufficiently inside the opamps the RF can be rectified in non-linear semiconductor junctions and cause dynamic internal bias level shifting. This type of distortion in not normally a consideration in data sheet information, but can be a practical issue in dac output stages.
In addition to the foregoing function of the output stage subsections, the output stage is also responsible for final reconstruction filtering, and hopefully for essentially complete removal of any residual RF noise.
Reconstruction filtering typically produces least audible artifacts if phase response is allowed to change only gradually (low group delay of that type). Frequency response of filter types that produce gradual phase changes in the audio band normally have very gradual frequency response roll off characteristics. For that reason, filter corner frequencies may be set to a few hundred kHz.
Of possible interest to any wishing to experiment with higher AVCC voltages and I/V opamp output swings while avoiding output cross over, LME49860 opamps are essentially LME49720 opamps rated to operate from as high as +-22v rails. With higher rail voltages, greater output swings are possible and is typically the case, opamp data sheet graphs show continued reduction in distortion as output swing is raised beyond what is possible using +-15v rails.
Anyway, output stages are typically comprised of subsections such as I/V and differential. They each bring their own issues. The I/V opamp runs with some common mode input offset, and typically is designed with a few small volts of output swing maybe 3v peak to peak. They also typically run into low impedances which might be too low for some output swings. Low impedances can help keep noise low, and low output swings can help keep the I/V opamps from getting into cross over where distortion rises. Graphs of opamp output distortion seem to be taken by manufacturers with fixed test signals and in part reflect the amount of time a test sine wave will spend in the cross over region. There are also graphs of distortion vs output load resistance which show low load resistance causes distortion. Both the graphs may be of limited applicability for opamps with limited output swings with output offset to avoid the cross over region.
While it may be possible to increase AVCC closer to maximum limits and gain a little benefit, it is not likely that is where most output stages produce most of their distortion.
I/V stages are typically followed by differential summing circuits that operate over greater common mode input ranges than the I/V stages, and they also remove DC offsets from the I/V stages so that the final dac output can be near ground level. Having the output near ground level can allow removal of output capacitors and or output transformers that would otherwise be required.
It could be that most first order analysis of output stage distortion occurs in the differential stage due to the less ideal input common mode offsets and output cross over. An additional source of apparent distortion in output stages appears to come from very low level RF leakage through the dac outputs. If amplified sufficiently inside the opamps the RF can be rectified in non-linear semiconductor junctions and cause dynamic internal bias level shifting. This type of distortion in not normally a consideration in data sheet information, but can be a practical issue in dac output stages.
In addition to the foregoing function of the output stage subsections, the output stage is also responsible for final reconstruction filtering, and hopefully for essentially complete removal of any residual RF noise.
Reconstruction filtering typically produces least audible artifacts if phase response is allowed to change only gradually (low group delay of that type). Frequency response of filter types that produce gradual phase changes in the audio band normally have very gradual frequency response roll off characteristics. For that reason, filter corner frequencies may be set to a few hundred kHz.
Of possible interest to any wishing to experiment with higher AVCC voltages and I/V opamp output swings while avoiding output cross over, LME49860 opamps are essentially LME49720 opamps rated to operate from as high as +-22v rails. With higher rail voltages, greater output swings are possible and is typically the case, opamp data sheet graphs show continued reduction in distortion as output swing is raised beyond what is possible using +-15v rails.
Last edited:
Ran out of editing time with the last post, but also wanted to point out that dacs are often run at very high sample rates. Some people want the dacs to produce frequency response commensurate with sample rates. Even with the highest sample rates there is probably no need to reproduce more than 88kHz to 96kHz. People who feel that lateral timing information and ultrasonic reproduction needs to be accurate may be satisfied with the above frequency response limits if channel phase information is accurately preserved.
Because output stage reconstruction filtering is fixed and the same for all sample rates, some filter frequency and phase characteristics must be chosen so as to provide for reproduction off all information that affects audibility of music, but that suppresses audibility of alias and other unwanted frequencies.
My leanings at this point are towards the rather old idea of a 4 or 5 pole 75kHz Gaussian filter, with some active and some passive poles. For a Gaussian response, some poles will necessarily be at higher frequencies. It may be that one pole will be lower. Filters always involve trade offs, but I think this is probably about right and would still be a good choice.
Because output stage reconstruction filtering is fixed and the same for all sample rates, some filter frequency and phase characteristics must be chosen so as to provide for reproduction off all information that affects audibility of music, but that suppresses audibility of alias and other unwanted frequencies.
My leanings at this point are towards the rather old idea of a 4 or 5 pole 75kHz Gaussian filter, with some active and some passive poles. For a Gaussian response, some poles will necessarily be at higher frequencies. It may be that one pole will be lower. Filters always involve trade offs, but I think this is probably about right and would still be a good choice.
Last edited:
Actually, that was about the same as my thinking when I modded the dac board I have right now. There are pictures in the thread giving a close up view of the layouts. The first iteration can be seen here: http://www.diyaudio.com/forums/digital-line-level/314935-es9038q2m-board-38.html#post5377834
More recent pictures with heat sinks added are also available, but the components under the heat sinks are the same as in the first picture.
What seems to be most important with these dacs is great attention applied to basics such as power quality, layout, ground plane usage, component linearity and value matching, etc.
Things that seem to matter much less include things like exact output filter pole frequencies and I/V output swing. They matter some, but you could change filter component values, some some of them by 50% or more and not clearly hear an audible difference if you only changed one in both channels. For others, maybe more critical.
As usual the idea is to figure out where most of the remaining SQ issues are coming from and focus attention there. Just because we know how to design and measure some things to very close tolerances doesn't mean they are what we should most be looking at. Just saying. I'm still trying to figure out what changes I might make to my existing modded dac to make the most difference, and I kind of suspect they are probably still in the power quality area. Maybe we will find out at some point.
EDIT: Anyway, I have taken a few stabs at attempting to respond to some of the issues you raise. Not sure if I have responded in a way that is helpful, or that has led to possibly less distance between us. Happy to continue the conversation if you would like.
EDIT 2: Also, looking at some of what you seem say, I'm not clear if we have been talking about all the same things. The dacs can be operated in voltage mode or current mode. Output voltage swing at the dac outputs is only an issue with voltage mode output. With current mode output, they are held at AVCC/2 and only the I/V opamp output swings. So long as we keep the swing down so as to avoid cross over distortion in the opamp, that is probably the limit we care about, right? (just checking)
Last edited:
Right, some of that occurred to me later: http://www.diyaudio.com/forums/digital-line-level/314935-es9038q2m-board-248.html#post5533245
Thing is we don't want the output voltage to follow fast changes in the reference voltage. The reference voltage output is filtered for that reason, and we only want the output voltage to regulate in response to changes in the load current. But we don't want AVCC load current changes to change the clock voltage, do we? I don't understand that.
Last edited:
some pics from mine that i promised some time ago
Thanks for the pics! Good to see how the project is coming along. Always curious, I hope it will be okay if I can ask a few questions:
Not sure if I recall if you are one of those interested in a through-hole output stage, or if you have something else in mind for that?
Also, any interest in the benefits of an Arduino to optimize a few things in the digital domain?