Hi everyone
Regardless of any other performance.
ES9018S refers max 135dB SNR in mono mode when all 8 channels are tied together. DSD1794 refers max 132dB SNR for 4 channels. I suppose both DACs are equal?
Regardless of any other performance.
ES9018S refers max 135dB SNR in mono mode when all 8 channels are tied together. DSD1794 refers max 132dB SNR for 4 channels. I suppose both DACs are equal?
That is only one spec. The two sound very different. What you like might be the result of implementation of personal taste. But saying based on the SNR spec from the datasheet they are equal is not my opinion. I like the 1794 better (in NOS implementation)
It would be like comparing cars based only on the top speed specification. Two cars with top speed of lets say 250kmh (possible in Germany 😉 ) can be TOTALLY different in driving style comfort whatever. Still personal. I prefer my Audi A8 over a Porsche for example - still others might feel completely different on that subject 😆
I realize you said regardless of any other performance. In that case both are very equal, as you will never need >130dB SNR
It would be like comparing cars based only on the top speed specification. Two cars with top speed of lets say 250kmh (possible in Germany 😉 ) can be TOTALLY different in driving style comfort whatever. Still personal. I prefer my Audi A8 over a Porsche for example - still others might feel completely different on that subject 😆
I realize you said regardless of any other performance. In that case both are very equal, as you will never need >130dB SNR
Just wondering if the two dac chips are 'equal' in how their noise manifests. There is some noise with no audio signal present. Noise may change some when an audio signal is present (depending on signal amplitude, etc). In some cases it may matter how noise is measured. A standard AP noise number is all that matters?
I can't comment on the DSD1794 but it looks like many ESS DACs have a "grainy" noise modulation at or below the RMS noise level -- and some don't manage to hide this in THD+N or IMD+N vs level plots (those are the ones with visible ESS hump in the AP test suites). The distortion residuals are pretty rough and jump around with the signal, this is visible either with large FFTs in the spectra or with a lot of averaging in the time domain. In DSD mode it appears to be less of an issue (but higher noise in general), though.
AKMs don't do that, and I would think the BB doesn't do that as well.
Whether any of this is audible?... I do have my doubts here...
Thanks Doede, 1794 has very good reputation. Regarding DACs, I have experience only with Wolfson WM8741 (a 10 year ago contract), and your opinion counts, and I have very good relations with TI 😉. I like very much the separate PCM-DSD inputs of 1794, though the cost is higher than ESS9018S. For 43.75 EU (Mouser) you also get the input multiplexer. Additionally, if your scope is to beat the 135 DNR of it, you should include dual 1794 so 30.77 X 2 = 61.50 EU (Mouser again) plus a DIX9211 or SRC4392. BTW do you think that upsampling (not oversampling) of e.g. an MP3 would be useful and audible?
Does it make any difference in performance the package? I haven't heard that again. To be sincere, the only difference I know is that the QFP64 is very difficult to solder manually yet with stencil and solder paste use. I feel safe with up to 48 pins (QFP48) when I assemble prototypes. Thanks for any further info.
Your original question was whether the DACs are equal, not whether they perform equally. They obviously are different: different designs from different manufacturers with different patents and different packages.
I haven't a clue how their performance compares.
I haven't a clue how their performance compares.
IME the cost of a dac chip is only a small fraction of the cost of building a high quality dac. Most of what affects SQ is all the circuitry surrounding the dac chip, not the chip itself. However the chip does set an upper limit on best possible performance. That said, IME most dac designs don't reach the chip's limit.
If you look more carefully at my first post, the original question was if 9018S and 1794 are equal regarding SNR (or DNR). So, no problem my good friend. Thanks a lot for your input and time.
I agree. Both manufacturers have a different approach regarding the DAC section itself. The cost matters only for e-bay sellers. On the other hand, managing 64 pins (and more) packages is quite difficult, yet during PCB layout design. You have to include 0402 - 0603 case resistors and capacitors, which makes it more difficult if you are going to assemble the PCB manually. So, the dual in line SSOP case is by far more convenient for us the DIY-ers 😉.
Yes, I have seen many. Personally, I like the method of stencil - solder paste. If you have the money to pay for a 2 - 4 layer prototype PCB (Chinese manufacturers charge 5 PCBs min, Eurocircuits charge 2 PCBs min, but the cost is the same) then the cost of stencil is just 8 - 10 EU. I place with fine tweezers all components, and then I place the PCB on a 3mm aluminum plate, and then on a kitchen hot plate. The trick is to remove the PCB for a while, when you see the first smoke of flux activation. Don't leave it to boil. Then place the PCB again on the hot plate, and wait until the tin is melted thoroughly. All parts are automatically aligned in place due to surface tension of melted tin. And, of course, are soldered at once. It is a rough approach to the solder paste curve given by manufacturers. However, SMT parts are a big headache. You should be very patient and calm down to work with them.
Last edited:
Maybe it's due to my Asperger's, but I couldn't and still can't derive that from your first post.
Anyway, Markw4 and KSTR have a point: sigma-delta DAC noise floors may or may not have all sorts of funny artefacts, so if you would amplify the DAC output signals so much that you would actually hear the noise, it would probably sound different. Besides, there are other performance aspects than just the noise.
Apparently, you don't suffer from Asperger's syndrome. As I noted above, the WM8741 is my only experience with DACs. It seems that you are more experienced than me on the subject. Now I'm going to build a more sophisticated digital project, including DAC, FM tuner, Bluetooth, Air PLay, USB 2.0, for myself. Since I have two SRC4392 and two DIX9211 samples available, could you advise if the upsampling that offers the SRC4392 is a good option?
Does your DAC need to synchronize (slave) to an incoming word clock, like it would when you have an S/PDIF, Toslink or AES3 input, or can your DAC be the clock master? In the first case, an advantage of asynchronous sample rate conversion is that you can have a nice and clean free-running DAC clock generator and use the ASRC to handle frequency differences with the clock of the signal source. In the second case, I don't really see the point.
In fact if you go for the SRC4392 or another asynchronous sample rate converter, I would recommend choosing output clock frequencies that have no relation with any of the usual audio sample rates. See section 5 of https://linearaudio.net/sites/linearaudio.net/files/03 Didden LA V13 mvdg.pdf for the rationale.
The SRC4392 is pretty good, but it has a few limitations. One is that it doesn't suppress imaging between 0.5 fs and 0.5438 fs much, but the same holds for almost all filters on board DAC chips. I find it inelegant, but whether it matters is questionable.
Another one is that its digital volume control is, according to a diagram in its datasheet, placed at the output. That means that reducing the volume by a few dB with the SRC4392's volume control won't solve intersample overshoot issues (clipping on peak sample normalized recordings). It won't matter if you already attenuate the signal somewhere before the SRC4392, on your computer for example.
If you only use the SRC4392 to sample up, you can switch off its decimation chain. That eliminates the passband ripple and other imperfections of the decimation chain.
In fact if you go for the SRC4392 or another asynchronous sample rate converter, I would recommend choosing output clock frequencies that have no relation with any of the usual audio sample rates. See section 5 of https://linearaudio.net/sites/linearaudio.net/files/03 Didden LA V13 mvdg.pdf for the rationale.
The SRC4392 is pretty good, but it has a few limitations. One is that it doesn't suppress imaging between 0.5 fs and 0.5438 fs much, but the same holds for almost all filters on board DAC chips. I find it inelegant, but whether it matters is questionable.
Another one is that its digital volume control is, according to a diagram in its datasheet, placed at the output. That means that reducing the volume by a few dB with the SRC4392's volume control won't solve intersample overshoot issues (clipping on peak sample normalized recordings). It won't matter if you already attenuate the signal somewhere before the SRC4392, on your computer for example.
If you only use the SRC4392 to sample up, you can switch off its decimation chain. That eliminates the passband ripple and other imperfections of the decimation chain.
SRC4392 is a good quality resampler. Upsampling for its own sake with an otherwise very well designed dac may only have a slight effect on subjective sound quality, if that would be the interest. However the chips are good for attenuating jitter of a SPDIF stream (EDIT: as I see Marcel already mentioned while I was still typing).
EDIT 2: Regarding Marcel's suggestion to use a non-standard reference clock frequency for an ASRC chip, when I looked into trying that I found it difficult to procure good quality oscillator modules similar to what is available at standard audio frequencies. Thus there may be a tradeoff to consider.
EDIT 2: Regarding Marcel's suggestion to use a non-standard reference clock frequency for an ASRC chip, when I looked into trying that I found it difficult to procure good quality oscillator modules similar to what is available at standard audio frequencies. Thus there may be a tradeoff to consider.
Hi Marcel,
Thanks a lot for the link. I read your very interesting and informative document, reading at the same time the document of SRC4392, DIX9211, the DIX4192 that you used in your project, PCM1792A, and of course some terminology on Wikipedia and other online sources. I also found a lot of resources stored on my PC since I implemented the WM8741 DAC. Unfortunately after 10 years, I remember almost nothing. By the way, I found in my stock a tray with six WM8741, as well as all the related external passive and active components. This makes me think of turning to Wolfson's because I only have one PCM1792A sample and no money left over right now. I have a prototype of the combo I designed then available and I am very pleased with the sound taste of the WM8741. So for discussion, the PCM1792A states a maximum DNR = 132dB, while the WM8741 128dB (both mono). Could I guess this difference is due to the built-in I/V converter? I'm sure the PCM1792A DNR at the output of the necessary NE5534 external I/V converter will drop to the same level. Probably the same goes for the much-honored ES9018S? Okay, that's all. I'm pretty sure I will continue with the SRC4392 and two WM8741 in mono mode. Since I have no idea about the different tastes offered by the many SRC4392 filters, plus those of WM8741, I will add more code, an LCD screen, and 4-5 "menu" buttons to my MCU to make experiments later. In fact, I'm going to build a DAC that will drive, with its native balanced outputs, a fully balanced 2X170W amplifier that I designed 2 years ago. Thanks again for the tips.
I don't know the answer to your question about the causes of the DNR differences between those DACs.
Regarding the SRC4392, it only gives you two filtering options: with or without decimation filtering. The (quite logical) recommendation from TI is to only disable the decimation filter when you don't need it, that is, when the output sample rate is greater than or equal to the input sample rate.
My DAC has several filtering options, but those are implemented in an FPGA module, not in the SRC4392.
Regarding the SRC4392, it only gives you two filtering options: with or without decimation filtering. The (quite logical) recommendation from TI is to only disable the decimation filter when you don't need it, that is, when the output sample rate is greater than or equal to the input sample rate.
My DAC has several filtering options, but those are implemented in an FPGA module, not in the SRC4392.