It’s fine the way it is now, the second differential stage can do any further filtering.
Hans
The AD829's compensation is external, I made it slow enough (with a cap) to work with the noise gain of the circuit, which was well above unity at HF. As it was custom compensated, yes in circuit it is faster than LT1028. I don't get what you mean by 'preclude' here?
It is quite possible that LT1028 sounded better due to some other reason than noise yes. So far I haven't been able to identify a potential candidate reason though, other than noise.
I inspected your file, and it's obvious it has zero headroom, the data is touching full scale.
So although I can imagine that some OS systems might go into overload,
on my PCM1792, that's OverSampling all content to 192Khz, it sounded terribly distorted in the original version but also exactly as worse in the -6dB version.
So IMO it's a victim of making sound as loud as possible just below clipping after having compressed the content.
Hans
Sure, this track suffers the "loudness" problem a lot! And it can take the rest of the system to its limits. However, my primary concern during this test was not distortion in general. I used it because it is revealing for what is happening close to full scale. At the part from 4:20 to 4:55, the vibrato of the trumpet is easy to follow even with my PC monitor's built in speakers, but only in NOS it can be felt as a live vibrating instrument. In other words, NOS seems to tolerate the loudness issue better than OS. With the attenuated version it just sounds ...compressed. Of course, this is my personal impression. Furthermore, I could detect this difference only with this kind of passages. Not with percussion even in this loud track.
As the number of low pass filtering stages increase, the requirement for fast response devices like the AD829 can become limited of advantage to other more dominant factors potentially bettered by slower devices like the LT1028. Good power supplies are considered a foundation to support advanced performance, whereupon PSRR and CMRR can also be considered as internal regulators that are dynamic in nature. Such regulation can be considered enhanced by laser trimming found in both the AD829 and LT1028. Under circumstances of lower bandwidths dynamic regulation (or PSRR) of the LT1028 is superior to the AD829.
It doesn't appear that noise can be considered as an independent cause supporting impact to sonics. For example, for signals being sonically reproduced at a 90dB SPL being then turned down to zero, any remaining noise cannot be heard if below the threshold of human hearing defined at 0dB SPL. This suggests that the absolute maximum of SNR required is conditional upon the SPL of sound being reproduced in relation to the threshold of human hearing at 0dB.
To gain some incite into SPL, consider what it means to target a 100dB SNR. If the threshold of human hearing is 0dB, being the equivalent to the sound of a mosquito at 3 meters in an anechoic room, this would support detectability up to 100dB, or the sound of a jack hammer. Hence if the sound of a jack hammer is turned off in an anechoic room the sound of the mosquito remains at the threshold in its absence. Although this satisfies the target to support 100dB SNR it doesn't satisfy its implications.
For 100dB to be of target benefit implies that the sound of the mosquito at 3 meters adversely affects the sound of the jack hammer (or other sounds) while producing the 100dB SPL at 1 meter. Hearing damage from long term exposure to sound is at 85dB, suggesting that an environment producing such levels is inherently unpleasant and whereupon the sound of the mosquito cannot be determined of unpleasantry if at 0dB. Ultimately this suggests that the physical reproduction of any noise can be ignored below 85dB and whereupon noise (as anything not signal) seems must be disturbing or corrupting the signal.
FWIW, I find the NOS PCM DAC when good enough to sound more dynamic.
It's more about power supplies for me. Nowadays everyone launching his DAC device is using integrated regulators with fresh EI not knowing so much about that. Of course it works... it just works
(often, but ok there are also very good modern DAC)
It's more about power supplies for me. Nowadays everyone launching his DAC device is using integrated regulators with fresh EI not knowing so much about that. Of course it works... it just works
(often, but ok there are also very good modern DAC)That type of argument shows up from time to time. IMHO it doesn't hold water. That's because it is based on the the assumption that human hearing and brain processing are linear, time invariant, and stationary. In reality hearing and brain processing satisfy none of those assumptions.
Last edited:
...It's more about power supplies for me. Nowadays everyone launching his DAC device is using integrated regulators with fresh EI not knowing so much about that. Of course it works... it just works
I believe that Frank (Lampie519) would agree with you about the prime importance of the power supply. He utilizes a rather interesting seeming high-frequency based supply arrangement, and without active regulation. You seem to be suggesting that you don't utilize active regulation either. Would you care to elaborate on the power supply scheme which you prefer?
Maybe dynamic isn't the right word, the nos sounds smoother. The AK os dac is no slouch and does sound very good, but my impression is that it's sharper, almost quicker on notes. Nos has all the same detail but presents them in a more relaxed manner. Voices sound more natural on nos.
I thought it would be a good idea to further examine the Hybrid I/V converter that recently popped up as an alternative.
Big advantage is that it acts like a passive I/V without the potential input oversteering of an active converter, but with almost the same extremely low noise of an active converter.
That’s why I tried to simulate a complete circuit, in this case projected on the popular PCM1792/1794, to get a better feeling for certain sensitivities.
The circuit in the first image shows a 5700R resistor, connected to the (+) input of U1/U2.
This resistor is used to the simulate the PCM1792/1794’s noise, but this input should be directly tied to gnd in a working solution.
In the plot connected to the circuit diagram, input voltages are shown as calculated in the Sim at U1/U2 and at the second stage, here an OPA1632, when offering a steep square wave current step from 2.3mA to 10.1mA, the maximum step size for this PCM1792/1794, a Dac that is supplying 6.2mA at rest.
Nothing dramatic happens and the amp behind the 10nF caps can handle this step without drama resulting in a smoothly dampened response with the chosen values for the caps.
Nevertheless, still a fast op-amp should be used for U1/U2, slew rate minimal 20V/usec and a GBW of around 100Mhz.
Because of the noise produced by the Dac, represented here by means of the 5700R resistor being almost 10nV/rtHz, it’s obvious that the noise requirement for U1/U2 is quite relaxed.
Up to 5nV/rtHz is no problem at all. So, a LT1468 is one of the many possibilities to be used here.
The second Image shows the Frequency Response, in red the output from U1+U2 and in blue the overall output from the OPA1632.
Output level in SE is 2.7 Vrms and in Diff twice as much with 5.4 Vrms.
Given the FR, it’s obvious that this I/V converter can be used both for OS and for NOS for all possible variations of Fs.
Third image shows the noise spectrum.
Noise ramps up at HF, still well under control higher in frequency, and results in ca 3.3uV from 20Hz to 20Khz as against a measured 2.7uV for a conventional active I/V converter using the same op-amps and gain.
All noise figures were taken from the diff output.
When A-weighting this noise, it reduces from 3.3uV to 2.4uV,.
So S/N calculates to 20*log(5.4V/3.3uV) = 124.2 dB or to 127.0 dB(A).
Hans
.
Big advantage is that it acts like a passive I/V without the potential input oversteering of an active converter, but with almost the same extremely low noise of an active converter.
That’s why I tried to simulate a complete circuit, in this case projected on the popular PCM1792/1794, to get a better feeling for certain sensitivities.
The circuit in the first image shows a 5700R resistor, connected to the (+) input of U1/U2.
This resistor is used to the simulate the PCM1792/1794’s noise, but this input should be directly tied to gnd in a working solution.
In the plot connected to the circuit diagram, input voltages are shown as calculated in the Sim at U1/U2 and at the second stage, here an OPA1632, when offering a steep square wave current step from 2.3mA to 10.1mA, the maximum step size for this PCM1792/1794, a Dac that is supplying 6.2mA at rest.
Nothing dramatic happens and the amp behind the 10nF caps can handle this step without drama resulting in a smoothly dampened response with the chosen values for the caps.
Nevertheless, still a fast op-amp should be used for U1/U2, slew rate minimal 20V/usec and a GBW of around 100Mhz.
Because of the noise produced by the Dac, represented here by means of the 5700R resistor being almost 10nV/rtHz, it’s obvious that the noise requirement for U1/U2 is quite relaxed.
Up to 5nV/rtHz is no problem at all. So, a LT1468 is one of the many possibilities to be used here.
The second Image shows the Frequency Response, in red the output from U1+U2 and in blue the overall output from the OPA1632.
Output level in SE is 2.7 Vrms and in Diff twice as much with 5.4 Vrms.
Given the FR, it’s obvious that this I/V converter can be used both for OS and for NOS for all possible variations of Fs.
Third image shows the noise spectrum.
Noise ramps up at HF, still well under control higher in frequency, and results in ca 3.3uV from 20Hz to 20Khz as against a measured 2.7uV for a conventional active I/V converter using the same op-amps and gain.
All noise figures were taken from the diff output.
When A-weighting this noise, it reduces from 3.3uV to 2.4uV,.
So S/N calculates to 20*log(5.4V/3.3uV) = 124.2 dB or to 127.0 dB(A).
Hans
.
Attachments
You treat the DAC as an ideal current source. Is its output impedance high enough to make that approach valid?
The obvious way to represent the DAC's noise is by a current source in parallel with the DAC output. One way to model that in SPICE, is to put a noisy resistor across the input of a current-controlled current source (preferably with one side grounded to prevent a floating cluster) and connect the controlled source's output across the DAC output. Why do you prefer a resistor in series with the op-amp's input?
The obvious way to represent the DAC's noise is by a current source in parallel with the DAC output. One way to model that in SPICE, is to put a noisy resistor across the input of a current-controlled current source (preferably with one side grounded to prevent a floating cluster) and connect the controlled source's output across the DAC output. Why do you prefer a resistor in series with the op-amp's input?
The resistor was just one way to add a noise source, free from interrupting the current and voltage from the Dac.
My first attempts where in the direction you suggested, but this final solution worked the most stable while varying other parameters.
The 5700R was determined by taking my existing Dac with a measured 2.7uV noise that's using these op-amps with exactly the same gain.
So, empirically I got this 5700R as the value to use to get exactly this 2.7uV.
I don't get your other question about the Dac's output impedance.
When it's able to drive a virtual input, it should even be more able to drive a 10nF cap IMO.
Maybe you can go in more detail to eventually add this to the Sim.
Hans
My first attempts where in the direction you suggested, but this final solution worked the most stable while varying other parameters.
The 5700R was determined by taking my existing Dac with a measured 2.7uV noise that's using these op-amps with exactly the same gain.
So, empirically I got this 5700R as the value to use to get exactly this 2.7uV.
I don't get your other question about the Dac's output impedance.
When it's able to drive a virtual input, it should even be more able to drive a 10nF cap IMO.
Maybe you can go in more detail to eventually add this to the Sim.
Hans
for the chip power supply : linear discrete, zener reg, emitter follower, Darlington with no feedback resistor at the output, R-Core traffo with RC, not fast diodes bridge, big attention made to passive components and final decoupling to enhance the transcient feeling with certainly a little touch of distorsion for the good sound. Nothing fancy, a classic without servo, oaps. works great with the best PCM chips : tda1541, PCM56P, AD1862... the first chip being complex for a good grounding in relation to his three voltages rails and decoupling. I don't know if everyday brand have time to experiment or just follow the white papers of the chip makers and focus themselves on the analog output stage only.
I dunno if my words makes sense as I have no tech background, just a diyer with ears. Sounding seems very good according the visitors though in direct benchmark with ambitious DACs.
I like what Rogic achieved with very few components : fast transcient power supplies while classic schemes. But I very don't know if quiet enough for modern Delta Sigma... with some tweaks, can go much beyond !
Last edited:
Ken,
May I suggest the following.
Let a number of contenders of this thread with a NOS take a 44.1 file of decent quality.
Now let them load this File in Audacity, convert it to higher fs, like 176Khz/24 which is still a multiple of 44.1 and 24 bit to keep the calculation errors under control.
Store it, load it again and downsample it to 44.1Khz/16 while adding dither.
Now compare this file to the original file to find out whether the sound still has the specific NOS sound.
The same can be done by going to 192/24 and back to 44.1/16, because this is no longer a multiple of 44.1 and probably needs even more processing depending on the used algorithms.
The beauty of doing this simple test is that you don’t need to modify hardware or compare multiple Dac’s, just a little bit of work on the computer that anybody can do while changing only just one variable.
It could give some more insight whether digitally manipulating content has a noticeable effect on the NOS sound, while still playing at the same 44.1Khz.
Hans
May I suggest the following.
Let a number of contenders of this thread with a NOS take a 44.1 file of decent quality.
Now let them load this File in Audacity, convert it to higher fs, like 176Khz/24 which is still a multiple of 44.1 and 24 bit to keep the calculation errors under control.
Store it, load it again and downsample it to 44.1Khz/16 while adding dither.
Now compare this file to the original file to find out whether the sound still has the specific NOS sound.
The same can be done by going to 192/24 and back to 44.1/16, because this is no longer a multiple of 44.1 and probably needs even more processing depending on the used algorithms.
The beauty of doing this simple test is that you don’t need to modify hardware or compare multiple Dac’s, just a little bit of work on the computer that anybody can do while changing only just one variable.
It could give some more insight whether digitally manipulating content has a noticeable effect on the NOS sound, while still playing at the same 44.1Khz.
Hans
Indeed. Mind you, you now have to reduce parameter R1 by a factor of four.
Regarding post #538, if you want to get hard clipping on intersample overshoots, you will have to export the interpolated signal in a fixed-point format like 24 bits, close Audacity, start it up again and import the fixed-point signal. That's because Audacity may use a floating-point internal format that can handle lots of overshoot.