AD826
FYI and Jan,
After some listening test this afternoon I added the AD826 to my list of favorite opamps. I think the AD827 indeed sounds a bit to sweet. The AD826 puts things more into perspective, making a "sharper" or more accurate soundstage; e.g. the places of the instruments are easier to point out. The price of this seems to be a slightly less wide or "airy" soundstage and less sweetness (what could be seen as negative because it 'sounds' better...). The AD826 sounds more honest I think.
When looking at the spec's: it has lower settling times and less of that 2nd harmonic. Also better rejection ratios, so all this seems to add up to the above. I couldn't find a single channel equivalent (yet) at AD.
Wouldn't it be nice to have a list of optimum spec's in the end and then start hunting for the perfect opamp??
Greetings,
Ray
FYI and Jan,
After some listening test this afternoon I added the AD826 to my list of favorite opamps. I think the AD827 indeed sounds a bit to sweet. The AD826 puts things more into perspective, making a "sharper" or more accurate soundstage; e.g. the places of the instruments are easier to point out. The price of this seems to be a slightly less wide or "airy" soundstage and less sweetness (what could be seen as negative because it 'sounds' better...). The AD826 sounds more honest I think.
When looking at the spec's: it has lower settling times and less of that 2nd harmonic. Also better rejection ratios, so all this seems to add up to the above. I couldn't find a single channel equivalent (yet) at AD.
Wouldn't it be nice to have a list of optimum spec's in the end and then start hunting for the perfect opamp??
Greetings,
Ray
Interesting. I didn't do any in-depth study, but cursory glance says the 826 is a lot like the 847. But the 827 is listed as dual 847. Strange....
I doubt that the risetime plays a big role though, no proof, just my gut feeling. I think the important specs are flat gain, flat distortion, flat PSRR (or almost so) over the audio band. Gain should not be too high, because high gain would give rise to more internal IM because of feedback. Lower distortion on the test bench with a singel or a couple of freqs, but with music, its a whole different ballgame.
Jan Didden
I doubt that the risetime plays a big role though, no proof, just my gut feeling. I think the important specs are flat gain, flat distortion, flat PSRR (or almost so) over the audio band. Gain should not be too high, because high gain would give rise to more internal IM because of feedback. Lower distortion on the test bench with a singel or a couple of freqs, but with music, its a whole different ballgame.
Jan Didden
Re: airy, flowery & lightfooted
Yeah, I know...
Believe me, I'm not a floating new age type of guy, it's just that you need SOME expression to describe...
I thought settling time was an important spec for post-DAC opamps.
The AD826 is about 30% faster than the 827. The other spec's look quite similar. So could it be the settling time is responsible for the differences I hear?
Anyone any thoughts on this? I would like to know more about the influence these spec's have on the actual sound-image instead of the image on an analyser.
Ray.
Bernhard said:airy, flowery & lightfooted
Yeah, I know...
Believe me, I'm not a floating new age type of guy, it's just that you need SOME expression to describe...
janneman said:
I thought settling time was an important spec for post-DAC opamps.
The AD826 is about 30% faster than the 827. The other spec's look quite similar. So could it be the settling time is responsible for the differences I hear?
Anyone any thoughts on this? I would like to know more about the influence these spec's have on the actual sound-image instead of the image on an analyser.
Ray.
Re: Re: airy, flowery & lightfooted
I thought settling time was an important spec for post-DAC opamps.
The AD826 is about 30% faster than the 827. The other spec's look quite similar. So could it be the settling time is responsible for the differences I hear?
--------------------------------------------------------------------
I think settling time and behaviour with capacitative load are very important.
Flatness means nothing.
I thought settling time was an important spec for post-DAC opamps.
The AD826 is about 30% faster than the 827. The other spec's look quite similar. So could it be the settling time is responsible for the differences I hear?
--------------------------------------------------------------------
I think settling time and behaviour with capacitative load are very important.
Flatness means nothing.
Well, settling time IS important if the signals going into the opamp are taxing it to the limits of it slew rate and settling time. Since the signals coming out of the DAC are already low-pass filtered, going from very fast settling to extremely fast settling would make a difference, since in both cases it is ample.
Now, flatness on the other hand is important. In fact, the problems with feedback is that the error signals being fed back are amplified by the full amp open loop gain, which normally is quite non-linear as well. That causes all kinds of intermodulation and harmonic distortion products. The feedback does make sure that the levels are low, but there will be a lot of signals harmonically unrelated to the music.
If you amp is linear to begin with, the absolute levels of the distortion may not be as low (if the open loop gain is low), but it will be more benign because there will be much less intermodulation.
I think in the end it will sound better. Our experiences with tube equipment, while not low in absolute distortion terms, but with less unharmonical components and less higher harmonics, point in the same direction. It is not very hi-fi in the absolute, literal sense, but may be more pleasant to listen to.
Jan Didden
Now, flatness on the other hand is important. In fact, the problems with feedback is that the error signals being fed back are amplified by the full amp open loop gain, which normally is quite non-linear as well. That causes all kinds of intermodulation and harmonic distortion products. The feedback does make sure that the levels are low, but there will be a lot of signals harmonically unrelated to the music.
If you amp is linear to begin with, the absolute levels of the distortion may not be as low (if the open loop gain is low), but it will be more benign because there will be much less intermodulation.
I think in the end it will sound better. Our experiences with tube equipment, while not low in absolute distortion terms, but with less unharmonical components and less higher harmonics, point in the same direction. It is not very hi-fi in the absolute, literal sense, but may be more pleasant to listen to.
Jan Didden
Konnichiwa,
Hmmm. Would you mind elaborating how the current output of a R2R DAC is already lowpass filtered? I sincerely hope you are not refering to the Digital Filter. This does not lowpass the current step on the output of the DAC at all.
Going from slow to fast MUST make a difference. Here is why. The DAC rearranges it's output current within a given time after the input data to the DAC has been read in and the aproriate "comit" input is set. The setteling time of the DAC is usually a small fraction of the actual sample rate, so a 96KHz DAC rated for 8 X Oversampling will settle it's current output in a small fraction of the 1.3uS sample period.
For arguments sake, let's assume the DAC settles in 1/100 of the sample period (really a bit slow for use without deglitcher) we have a currentstep with a 13nS timespan and at least theoretically this current step may be from zero to fullscale. Now a 13nS Period is equal to 76MHz.....
Can't for the life of me see why. In I/V applications where the Op-Amp is operating into a RC Load (feedback loop) and attempts to keep the voltage on the inverting input at Zero. Settling time of the Op-Amp itself is quite crucial, if it settles to slowly it will not follow the current step accuratly.
The OPA627 settles in 550nS, the OPA604 in 1000nS while the AD811 settles in 25nS and the AD8065 settles in 65nS (all to 0.1%). I could throw in a few more but I have observed a strong correlation between settling time and sound quality in I/V applications.
The often used band aide of a capacitor between the inverting input and ground is not very smart either.
Sayonara
janneman said:
Hmmm. Would you mind elaborating how the current output of a R2R DAC is already lowpass filtered? I sincerely hope you are not refering to the Digital Filter. This does not lowpass the current step on the output of the DAC at all.
janneman said:
Going from slow to fast MUST make a difference. Here is why. The DAC rearranges it's output current within a given time after the input data to the DAC has been read in and the aproriate "comit" input is set. The setteling time of the DAC is usually a small fraction of the actual sample rate, so a 96KHz DAC rated for 8 X Oversampling will settle it's current output in a small fraction of the 1.3uS sample period.
For arguments sake, let's assume the DAC settles in 1/100 of the sample period (really a bit slow for use without deglitcher) we have a currentstep with a 13nS timespan and at least theoretically this current step may be from zero to fullscale. Now a 13nS Period is equal to 76MHz.....
janneman said:
Can't for the life of me see why. In I/V applications where the Op-Amp is operating into a RC Load (feedback loop) and attempts to keep the voltage on the inverting input at Zero. Settling time of the Op-Amp itself is quite crucial, if it settles to slowly it will not follow the current step accuratly.
The OPA627 settles in 550nS, the OPA604 in 1000nS while the AD811 settles in 25nS and the AD8065 settles in 65nS (all to 0.1%). I could throw in a few more but I have observed a strong correlation between settling time and sound quality in I/V applications.
The often used band aide of a capacitor between the inverting input and ground is not very smart either.
Sayonara
janneman said:
Would that be the case for my SM5872 DAC, since this DAC has a 'raw' unfiltered output, so this DAC would 'need' a low settling time opamp more than a chip with built-in opamps or filter (like TDA1547, PCM1710...).
I agree. The open loop design should be as stable and lineair as possible.
Better to not use FB at all. The signal always arrives too late! And if you use it to 'improve' the spec's of a bad design it will sound bad anyhow. Low levels of distortion by means of lots of feedback are fake.In my tube amp I've applied only 6dB of FB to control the speaker a bit better at low frequencies. If I make that 12...15dB the sound becomes dull, so this also can happen with an opamp it seems.
Greetings,
Ray.
Hi KYW
Well, isn't the DAC output indeed low-pass filtered? I mean the internal (digital) filter. If it doesn't slow down the current step rise time, how does it then filter lo-pass? And the cap between opamp output and input means that for high frequencies (remaining current step) the opamp requires much less output amplitude to satisfy the requirement to null the input. That relaxes the slew rate & settling time considerably, in my view.
As for amplifier flatness BEFORE feedback, I don't know how to express my view better than I did. There is some supporting documentation, which I haven't handy at the moment unfortunately.
I agree that the cap from inv input to gnd is a bad idea. I have experimented with a small inductor in series with the DAC current output, though. While it filters the I/V input current even more from hf components, it makes them (the hf components) appear at the DAC output. I didn't do any listening tests, but it looked ugly enough not to pursue it any further. Do you have an opinion on that?
Jan Didden
Well, isn't the DAC output indeed low-pass filtered? I mean the internal (digital) filter. If it doesn't slow down the current step rise time, how does it then filter lo-pass? And the cap between opamp output and input means that for high frequencies (remaining current step) the opamp requires much less output amplitude to satisfy the requirement to null the input. That relaxes the slew rate & settling time considerably, in my view.
As for amplifier flatness BEFORE feedback, I don't know how to express my view better than I did. There is some supporting documentation, which I haven't handy at the moment unfortunately.
I agree that the cap from inv input to gnd is a bad idea. I have experimented with a small inductor in series with the DAC current output, though. While it filters the I/V input current even more from hf components, it makes them (the hf components) appear at the DAC output. I didn't do any listening tests, but it looked ugly enough not to pursue it any further. Do you have an opinion on that?
Jan Didden
janneman said:
I put the Dual AD-827 in my Magnavox CDP that uses the TDA1541 DAC and a Dual OPA as an I/v Converter and Output Buffer. And I liked the sound so much that the AD827 stayed installed for over 7 years.
I thought about the AD827 again while playing with this DAC
http://www.aoselectronics.com/pace.html
I replaced the LM6172 and also tried the OPA627 the AD-8620 and the AD8066 so far it’s a toss up between the OPA-627 and AD827. Each is warm sounding yet the attack is sharper with the AD827. Now i think the OPA627 may have a more pleasant and friendly presentation yet as noted by others on this thread the AD8827 has better sound stage. The AD8066 gives the lowest Distortion yet lacks Bass extension. The resolving powers of the AD8066 are its hall mark this OPA let you hear deep into the recording to the point that in some recordings you can hear people talking in a whisper and tell where they are in relation to the performers quite a nice thing to experience. The OPA627 has the extension however seams to lack control giving a somewhat fat sound to the mid-bass, never the less the OPA627 is real musical in this application and while the requirements for an op amp after the DAC in the TDA1541 and the PCM1793 as used in the aos Electronics DAC. I am also going to try an AD826 and LT1028 or LT1128 depending upon if the LT1028 will remain stable at unity gain in this DAC.
By the way jan i agree with your assesment of wide open loop bandwidth and flat PSRR, CMRR,THD and Open loop bandwidth beyound 10KHz.
janneman said:
I think you are mixing things up here Jan. A digital filter affects
the sample values. The DAC still has the same rise and fall
time. It doesn't know if the sample values have been altered
by a filter or not. Now, the difference between subsequent
sample values will be smaller with a digital LP filter, but that
is something else.
ppl said:
This is not unusual, as soon as you clean up the mid/high often subjectively it sounds like you have less bass. Did you test this with material that has real low bass, not 50-80Hz, but 30Hz and below? There is not much material out there that has it. We are quite accustomed to what we think is low bass but in reality is above 50Hz or even above 100Hz. It may *seem* to drop away after cleanup of the midrange.
Jan Didden
Jan,
the filtering, and thus attenaution of HF, is in the digital domain.
Think of it like this very simplified example. Say that we have
three samples coming from the CD with values 0, 500 and
1000. If we don't have a digital filter, the DAC will output these
values, within the accuracy possible. The rise/fall time between
samples will be, or at least should be, very small compared
to the length of a sample. It will probably be somewhat slower
for larger steps, which causes an error, but if it is fast enough
we assume it can be ignored.
Now, assume we have a digital filter. This filter process the
digital data and is not part of the DAC (but could be on the
same chip, of course). While an analog filter will tend to reduce
the slew rate of the signal, a digital filter will tend to reduce
the difference between adjacent samples, since the closest
we can come to something like slew rate in the digital domain
is to look at this difference. In the example above we might
get maybe the sample values 0, 400 and 850 (or whatever,
depending on the filter). These values will go to the DAC
and be output, but the rise and fall time for changing output
values of the DAC will still be the same, or at least the DAC
will change its output with the same slew rate in both cases.
Somewhere after the DAC you need an analog LP filter to
get rid of the HF components caused by the squarishness of
the sample pulses, but that is something you need in both
cases. When the signal has been prefiltered in the digital
domain, you will never get as large differences between
samples as in the unfiltered case and hence you get
slightly less HF in the spectrum.
the filtering, and thus attenaution of HF, is in the digital domain.
Think of it like this very simplified example. Say that we have
three samples coming from the CD with values 0, 500 and
1000. If we don't have a digital filter, the DAC will output these
values, within the accuracy possible. The rise/fall time between
samples will be, or at least should be, very small compared
to the length of a sample. It will probably be somewhat slower
for larger steps, which causes an error, but if it is fast enough
we assume it can be ignored.
Now, assume we have a digital filter. This filter process the
digital data and is not part of the DAC (but could be on the
same chip, of course). While an analog filter will tend to reduce
the slew rate of the signal, a digital filter will tend to reduce
the difference between adjacent samples, since the closest
we can come to something like slew rate in the digital domain
is to look at this difference. In the example above we might
get maybe the sample values 0, 400 and 850 (or whatever,
depending on the filter). These values will go to the DAC
and be output, but the rise and fall time for changing output
values of the DAC will still be the same, or at least the DAC
will change its output with the same slew rate in both cases.
Somewhere after the DAC you need an analog LP filter to
get rid of the HF components caused by the squarishness of
the sample pulses, but that is something you need in both
cases. When the signal has been prefiltered in the digital
domain, you will never get as large differences between
samples as in the unfiltered case and hence you get
slightly less HF in the spectrum.
Re: Re: Re: airy, flowery & lightfooted
Jan and Christer, I think you are both saying almost the same thing here, and I agree with both of you. I made plot in Matlab to illustrate this.
The blue signal is a 20kHz sinewave sampled at 44100. The green curve is the same sinewave sampled at 8*44100. Then a zero-order hold is applied to both signals as in a R-2R DAC. By oversampling with a bandlimiting digital filter you eliminate the chance of the signal changing from full-scale positive to negative, thus reducing the step-size. However the sample period is also reduced, and without doing any heavy math I think the step-size is reduced by the same factor as the sample-period (in this case 8) so that the slew-rate requirements stay the same for both cases!
However one might argue that a slower slew-rate would just act as a low-pass smoothing filter, but we want to filter the signal with passive RC networks and not by transistor limitations don't we?
janneman said:
Christer said:
Jan and Christer, I think you are both saying almost the same thing here, and I agree with both of you. I made plot in Matlab to illustrate this.
The blue signal is a 20kHz sinewave sampled at 44100. The green curve is the same sinewave sampled at 8*44100. Then a zero-order hold is applied to both signals as in a R-2R DAC. By oversampling with a bandlimiting digital filter you eliminate the chance of the signal changing from full-scale positive to negative, thus reducing the step-size. However the sample period is also reduced, and without doing any heavy math I think the step-size is reduced by the same factor as the sample-period (in this case 8) so that the slew-rate requirements stay the same for both cases!
However one might argue that a slower slew-rate would just act as a low-pass smoothing filter, but we want to filter the signal with passive RC networks and not by transistor limitations don't we?
Attachments
Also a very interesting paper here (as posted in a different thread), where the author demonstrates that slew-rate induced distortion is the same as jitter distortion and then compares two op-amps (pages 11-18) one with 500v/us and one with 50V/us for different sampling rates. Some real eye-openers here (for me at least...) Notice how the non-os case is much more sensitive to slew-rate than the higher sampling rates.
and then compares two op-amps (pages 11-18) one with 500v/us and one with 50V/us for different sampling rates. Some real eye-openers here (for me at least...) Notice how the non-os case is much more sensitive to slew-rate than the higher sampling rates.ojg said:Also a very interesting paper here (as posted in a different thread), where the author demonstrates that slew-rate induced distortion is the same as jitter distortion
Interesting, but not surprising now when you say it. Jitter is an
error in the time domain that manifests itself as amplitude
error after filtering. Limited slew rate also gives an error in
the time domain, but also, I think, simultaneously in the
amplitude domain. It sounds intuitively reasonable that they
should be very similar at least. Paper in printer. Will read when
time permits.
Konnichiwa,
First, the PCM1704 in the example has not got a digital filter.
I repeat, with proper DAC's (eg R2R and related, not the "delta sigma dreck) you get an output where with each sample period there is a different output current which settles to nominal within a small fraction of the sample period.
Now delta sigma DAC's are a very different kettle of fish, however their use of agressive noiseshaping and very high oversampling ratio's usually makes their output's even nastier in terms of slew rate limiting / setteling time which is bad news if the manufacturer decided to cripple the DAC sonically by putting a crappy Op-Amp on board which will be slewing. Of course it's good news for the usual hamfisted and brainless designers of consumer gear, at least they cannot get it wrong anymore.
It adjust the digital data in such a way that the resultant freqency response is altered. This happens purely to the datavalues which afterwards are converted to current values (with steep steps inbetween) for "Multibit" DAC's or to noise pulses (current or voltage - with a really highly extending noise spectrum) for Delta Sigma DAC's.
Voltage - yes, current - no. You must avoid to think always in the voltage domain.
A DAC with a 20 Bit notinal dynamic range and 8 X Oversampling at 96KHz needs to settle VERY quickly. The Op-Amp in theory at least should settle faster than the DAC to not limits it's performance. Again it is a BBFOTVO to see that setteling is the end it all in an I/V converter.
Ideally a non-looped design is used which in effect settles immediatly, with zero delay.
A wide bandwidth with low open loop gain is a sideeffect of making a high slewrate Op-Amp. However as we never want the loop around the I/V converter to "open" the open loop behaviour matters only in so far as it supports the short seteling time and thus keeps the feedloop "locked".
I prefer to keep I/V conversion either open loop or passive, so to me neither matters, unless I do "compatible" mods on commercial gear. There keeping the AC on the DAC as small as possible always sounds better, why, that is another question.
Of course not. How would it do that?
The digital filter adresses the signal ONLY. The fast risetime current pulses are part of the conversion mechanism of the DAC and occour AFTER the digital filter.
Sayonara
janneman said:
First, the PCM1704 in the example has not got a digital filter.
I repeat, with proper DAC's (eg R2R and related, not the "delta sigma dreck) you get an output where with each sample period there is a different output current which settles to nominal within a small fraction of the sample period.
Now delta sigma DAC's are a very different kettle of fish, however their use of agressive noiseshaping and very high oversampling ratio's usually makes their output's even nastier in terms of slew rate limiting / setteling time which is bad news if the manufacturer decided to cripple the DAC sonically by putting a crappy Op-Amp on board which will be slewing. Of course it's good news for the usual hamfisted and brainless designers of consumer gear, at least they cannot get it wrong anymore.
janneman said:
It adjust the digital data in such a way that the resultant freqency response is altered. This happens purely to the datavalues which afterwards are converted to current values (with steep steps inbetween) for "Multibit" DAC's or to noise pulses (current or voltage - with a really highly extending noise spectrum) for Delta Sigma DAC's.
janneman said:
Voltage - yes, current - no. You must avoid to think always in the voltage domain.
janneman said:
A DAC with a 20 Bit notinal dynamic range and 8 X Oversampling at 96KHz needs to settle VERY quickly. The Op-Amp in theory at least should settle faster than the DAC to not limits it's performance. Again it is a BBFOTVO to see that setteling is the end it all in an I/V converter.
Ideally a non-looped design is used which in effect settles immediatly, with zero delay.
janneman said:
A wide bandwidth with low open loop gain is a sideeffect of making a high slewrate Op-Amp. However as we never want the loop around the I/V converter to "open" the open loop behaviour matters only in so far as it supports the short seteling time and thus keeps the feedloop "locked".
janneman said:
I prefer to keep I/V conversion either open loop or passive, so to me neither matters, unless I do "compatible" mods on commercial gear. There keeping the AC on the DAC as small as possible always sounds better, why, that is another question.
janneman said:
Of course not. How would it do that?
janneman said:
The digital filter adresses the signal ONLY. The fast risetime current pulses are part of the conversion mechanism of the DAC and occour AFTER the digital filter.
Sayonara