ES9018 vs PCM1704UK

[XAHTP]

All else being equal, how do these chips compare? Which one is more suitable for USB implementation?

 

[barrows]

not a possible comparison, as these DAC chips are very different, and have entirely different needs for the circuitry around them to make a complete DAC.

The 1704 is a DAC only, and requires a separate digital receiver of some type, as well as a digital filter. then, it also requires a very good I/V stage (specifically tuned to its needs) and an output stage.

The ESS 9018 is a very sophisticated, inclusive DAC chip. It features an innovative onboard SPDIF receiver, a unique upsampling/oversampling method, and programmable digital filters. The ESS can be used in either voltage or current output mode, and requires a very specific following output stage to achieve maximum performance.

Oh yeah, the 1704 is a mono chip, so a DAC needs at least two for stereo, and the ESS 9018 is an 8 channel chip, which uses 4 DACs per channel in stereo mode for lower noise/distortion.

Since these chips cannot be used in an "equal" way, it is impossible to make direct comparisons, one can only make comparisons of entire DACs which incorporate these chips.

Either one will work fine for a USB DAC, but the performance on USB will be determined by the performance of the USB receiver,a nd this will have nothing to do with the DAC chip.

The specs on the ESS 9018 are state of the art: by the numbers it outperforms anything else, and in my experience, when properly implemented (which is not easy) it offers very, very good sonic performance

 

[PeterSt]

... or from another perspective ...

 

One ESS chips will cost around $40 (maybe 10 less, maybe 10 more) while one 1704U-K costs also that if you buy a 100.

 

So how many go into an 8 channel DAC ?

1 ESS and 8 1704's. Haha.

 

But the major difference between the two is that the ESS is of the (1 bit) oversampling type, while the 1704 is 24 bit multibit and this sure has some impact.

 

It has impact on measuring because no OS chip can be measured for its real merits (only periodic waves can be measured while the oversampling and filtering algorithms highly depend on just that (periodic-ish waves) which doesn't exist really in music) - while an NOS chip shows the real figures (so this is 0.00002% THD+N or so for the ESS and 0.0008% for the 1704U-K, while in practice ? ...

 

Sonically I prefer the 1704, but first you have to die a 1000 deaths in order to get that I/V right.

 

I'll advise everybody to buy the ESS, so maybe the price of the 1704 drops. :-)))

 

Better read Barrows' post, he seems to be more serious than I am. Although I try to be too.

 

Peter

 

[Miska]

It is pretty much useless to compare just DAC chips, since performance of the overall system depends so much on the data fed to the converter stage as well as analog output stages. And what matters in the end is performance of the complete system.

 

Both of these chips are very good when used properly, and there are other equally good ones.

 

There are various external factors affecting the performance...

- Power supplies

- Clock jitter

- I/V stage

- Output buffer

- etc..

And especially in case of PCM1704 since it depends more on external components on input and output side:

- Properties of the preceding oversampling digital filter; like transition shape, amount pre-ringing, amount of stop-band leakage, etc. (usually assumed part for this is DF1706)

- Analog reconstruction filter

Above two define the frequency response, phase response, transient performance and total amount of spectral leakage above fs/2 which define how well the analog waveform is reconstructed.

 

From USB point of view the only difference is that ES9018 being high frequency delta-sigma converter it is less sensitive to clock jitter than the low frequency PCM converter PCM1704.

 

[Panelhead]

I look at the Metric Halo gear as an example. The latest products use AK4395 dacs. These are far from state of the art chips. Yet all I hear is very positive response to the sound quality.

Many still cling to TDA1541 chips. The results are dependent on the weakest link in the unit. Almost any high quality dac is good enough if it is the weakest link.

 

George

 

[PeterSt]

George - At first I wanted to let know how good your post actually is, and/but that it nicely shows how much more can be done to improve on sound. But then I saw that I read it the other way around from what you wrote, and it looks like you also wrote it the other way around from what you wanted to say ...

 

If the chip is the weakest link (which is what you said) it will determine SQ directly. But if outboard stuff is the weakest link, and DAC chips outbetter that by far, the chip doesn't matter much (which I think is what you meant to say ?).

 

Maybe it is allowed to translate it to something like this (based upon my own experience) :

 

If a chip shows 0.0008 % THD+N it would be my goal to achieve that same number at the net output. You can also say it is the most difficult to reach the 0.0020 % I did, measured at the end of the interlink.

Now, theoretically, if the chip shows e.g. 0.0040% to begin with, I had to do less to reach that figure - hence in my case I would have met the chips specs. Nothing left to improve on it, and much easier to achieve. Thus also faster achieved.

 

Blahblah

Peter

 

[Panelhead]

What I was thinking, and maybe not writing, is if the layout, power supplies, output stage, input receiver, and drivers all are integrated to level where the dac is the weakest link the results will be excellent.

A better example of this is Gordon's Wavelength dacs. Those 1543 chips are way down the most people's list. Yet owners love the dacs. The rest of the package is dialed in to let these old, low cost chips shine.

I looked over a listing recently of the chips used in most the the prosumer interfaces. There must have been 15 different ADC's and 20 different DAC chips. And no correlation of sound quality. Some expensive gear used cheap chips and some cheap used top line chips.

A USB0404 interface uses the same chip as a ULN8. Guess which one I want?

 

George

 

[Miska]

Based on my experience most of the cost for ADC/DAC comes from the analog stages and power supply, and at least for small quantities also from PCB and casing.

 

I'd guess large part of the cost difference between pro gear and high-end comes just from the casing. Most pro people are just fine with normal front panel, no need for 1 cm thick machined aluminium piece and fancy buttons/knobs.

 

 

[regal]

I wouldn't say chip cost is insignificant any longer. The last batch of PCM1704k's sold out at over $70 each. The new batch is priced higher.

 

If we look at probably BB's last reference board (the pcm1702 board) we see the BB engineers used 4 DAC chips per channel all in parallel. Its no secret that the PCM1704 sounds better in multiples (they ain't pcm63's). So to do per reference a balanced PCM1704 would be 16 chips, but probably 8 is good even balanced. So we have $80x8= $640 just for the DAC chips. No other component, not even the case comes to this price in any DAC. The only higher cost are the fixed ones like the engineering and programming.

 

 

I just looked at a -60dB 1khz FFT of a ESS9018 (Yulong pdf) and it showed no harmonics, quite amazing. I have not had much luck with switch mode DAC's, and their actual distortion 0-20MHZ is over 1% THD, but the R2R's are becoming very expensive and the output stage is such a PIA.

 

If I hold onto these 8 PCM1704k, will you promise to build me a Phasure when my kid grown up and out of the house:)

 

 

[PeterSt]

Yea, I saw that just today. It looks like they don't want to sell them anymore (and thus they can really stop production).

 

For fun : The Phasure NOS1's design allows to run with 16 d/a chips for two channels (differentially and parallelled). But honestly, I never tried it myself, no matter how easy it would be (it's just dipswitches stuff).

 

I don't know how long it takes until your son goes off with some nice girl, but maybe by the time I have a few left myself. Or we make a 16 together with yours ... :-)

 

Regards,

Peter

 

PS: You really got me thinking ... why don't I really try the 16 ? THD+N of 0.0018% should go down to 0.0009% ... But this is from the paralled theory only; there is more; So, I can already see that it will be more close to 0.0003% or something like that, PLUS that's with 192KHz output only (I can't measure more), while it really does 768Khz ...

Audible ? no idea ...

 

[Miska]

I have not had much luck with switch mode DAC's, and their actual distortion 0-20MHZ is over 1% THD

 

Where on earth do you get that kind of figures from?

 

That's nothing like what I've ever seen from a properly designed DAC.

 

 

[KingStyles]

i have always wondered what the difference is between the 1704k and the 1704u-k is?

 

[PeterSt]

It is different. There's the PCM1704U and the PCM1704U-K. The "K" grade has better THD specs.

 

Peter

 

[Miska]

A USB0404 interface uses the same chip as a ULN8.

 

Just for fun, here's plot of 1 kHz played at -60 dBFS through E-MU 0404 USB DAC and looped back into it's instrument input ADC and analyzed. Fourth harmonic is highest in level, just below -140 dbFS. I calculated THD to be roughly 0.01% at -60 dBFS level from the two visible harmonics (third and fourth).

 

 

 

[regal]

i Where on earth do you get that kind of figures from?

 

i That's nothing like what I've ever seen from a properly designed DAC.

 

 

Noticed I said 0-20mhz noy khz, if you understood how S-D DAC's work you would have gotten the joke.

 

[Miska]

Noticed I said 0-20mhz noy khz, if you understood how S-D DAC's work you would have gotten the joke.

 

Of course I do understand since I have my own delta-sigma modulators, even included in the player.

 

I have naturally run spectrum analysis on the DAC output and can tell that there's nothing beyond 250 kHz.

 

Have you ever checked how multibit DAC output looks like above half of the sampling frequency?

 

I would also like to hear why especially THD would be so bad? SDMs give especially low THD, because all the high frequency products are practically pure noise, uncorrelated with the signal. Unlike multibit ladder DACs... Since THD is Total Harmonic Distortion and there are no correlated harmonics in SDM output it kind of doesn't make sense in general.

 

 

[PeterSt]

Have you ever checked how multibit DAC output looks like above half of the sampling frequency?

 

Eh, clean ?

When properly filtered of course.

 

Or you know more than I know/see.

 

?

 

[Miska]

Eh, clean ? When properly filtered of course.

 

Those ones that have high enough oversampling and good enough digital filters in the oversampling, so that the band between original and new fs/2 is clean too.

 

Of course the analog filter needs to be good too, to have 140 dB attenuation by the frequency where images hit.

 

I've seen much more leaky multibit DACs than leaky SDM DACs. And that leakiness is much worse because it is correlated.

 

 

[regal]

i I've seen much more leaky multibit DACs than leaky SDM DACs. And that leakiness is much worse because it is correlated

 

 

And you have a 12Ghz scope ? Look it was just a joke