Measurements of an ES9023 DAC

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Hi Gaetan,

Good question. I do not know why it has such a big impact on the sound as I do not know how they implemented the function in the chip.

If you want to test it add a 121K resistor from pin 6 to ground. This reduced the output in my design to ~2V RMS down from just under 2.1V RMS without it while using 3.6V rails for the 9023. It was night and day in my layout.

Dave
 
I'll add one more suggestion - DO NOT use the limiting resistor. While it may reduce clipping for signals we likely never will see in music, it's use is very detrimental to the sound.


Dave

Sorry late reply, thanks Dave for your recommendations.

I'm testing different caps in my ess9023 dac, all ceramics, all Oscons with small esr ceramic on Cfly. found the latter very good. still need to test good film caps, which one are you using?

I want to draw the attention to one thing I' ve discovered that brings the sound to a new level, stereo image gets wider, natural sound.. very good sound.

I wanted to get rid of the internal charge pump inverter, so I omitted the Cfly capacitor (the capacitor between CP and Cn), thus disabling de facto the switched transfer of current from Vin to NEG.
Since negative voltage is not coming from the pump, I tried connecting a quiet good source (lifepo) to NEG pin.. actually a lifepo with the minus to NEG and positive to GND.

DC offset similar to the original circuit.. 2/3 mV.
DC servo seems to track well the -3.2/3.3 from the lifepo.. the pump circuit output was a bit less (-3.1V if I remember correctly).

in my setup very good result with this simple mod.

anyone willing to try and confirm this?

PS: pay attention to DC offset.. if you leave connected alone Vneg you'll find DC at the output. you need both Vin, NEG, and omit only Cfly across Cp/Cn.

Thanks
regards
Valeriano
 
Last edited:
That is an interesting observation.
A brave one as well, or you have the internal schematics ??

;)

Congratulations,
Patrick

Thanks Patrick!

unfortunately I have only access to the datasheet which is telling nothing about internal operation of the circuit.

I got some information about charge pump inverters (switched capacitor) using datasheet from TI and others.. since Cfly is vital to the charge transfer from Vin to NEG, I asked myself why not trying omitting this capacitor.. and directly source the negative voltage to NEG pin.

some additional info, charge pump internal oscillator is connected to the system clock, without clock pump is not working.
Vreg on the other hand is generated from Vcc regardless of the clock connected.. so no switching related operation on this reference.

I was concerned about spurious 1.5Mhz coming from the pump even without Cfly.. I don't see anything with my cheap Rigol scope.

regards
Valeriano
 
Disabled Account
Joined 2002
Patrick, did you try Wima MKS2 in 2.5 mm pitch yet ? I wanted to try that one but my ES9022 DAC did not survive the many mods and soldering. Very frustrating to build something that dies but the last mod was the final. Anyway, I could not judge the Wima anymore (well, the DAC distorted terribly as the chip was fried) but it seems a very good candidate.
 
Last edited:
Disabled Account
Joined 2002
Sorry late reply, thanks Dave for your recommendations.

I'm testing different caps in my ess9023 dac, all ceramics, all Oscons with small esr ceramic on Cfly. found the latter very good. still need to test good film caps, which one are you using?

I want to draw the attention to one thing I' ve discovered that brings the sound to a new level, stereo image gets wider, natural sound.. very good sound.

I wanted to get rid of the internal charge pump inverter, so I omitted the Cfly capacitor (the capacitor between CP and Cn), thus disabling de facto the switched transfer of current from Vin to NEG.
Since negative voltage is not coming from the pump, I tried connecting a quiet good source (lifepo) to NEG pin.. actually a lifepo with the minus to NEG and positive to GND.

DC offset similar to the original circuit.. 2/3 mV.
DC servo seems to track well the -3.2/3.3 from the lifepo.. the pump circuit output was a bit less (-3.1V if I remember correctly).

in my setup very good result with this simple mod.

anyone willing to try and confirm this?

PS: pay attention to DC offset.. if you leave connected alone Vneg you'll find DC at the output. you need both Vin, NEG, and omit only Cfly across Cp/Cn.

Thanks
regards
Valeriano

What a great idea ! I hope a negative power supply to that pin will make such a difference too. And will the chip survive this eventually ?
 
Last edited:
What a great idea ! I hope a negative power supply to that pin will make such a difference too. And will the chip survive this eventually ?

Thanks Jean Paul,

in my final setup I plan to use low voltage salas for positive and negative supply, I'll report back soon.
Hopefully it will survive, I see no reason why should be different than a voltage provided by a battery.

regards
Valeriano
 
The charge pump will not ease working even without Cfly.
I need to try this some time, though we also have no complaints about using the charge pump, if you choose the right caps.
Try some SMD film caps to compare with the MLCCs.


Patrick

sure, it sounds good with MLCCs.
I'll try SMD film caps, too.

I'll be happy if you try negative supply without Cfly :)
thanks
Valeriano
 
Sorry to bump an old thread, but there never seemed to be a decent conclusion on how to use a symmetrical power supply with the ES9023.

A few weeks ago I tried it for my self, using a dual opamp power supply with 3.6V zener diodes. (Used low noise Maxim MAX9632) I connected the negative supply directly on the "NEG" pin, and left the chargepump pins floating. No 1.5MHz noise can be measured on the analog output when the chargepump pins are disconnected.

When the positive supply is not connected, the resistance between NEG and GND is very low, because my opamp supply is limited at roughly 90mA, nothing burned. Otherwise, make sure to turn on negative supply after positive supply!!!

Dual supply is an improvement over a single supply ES9023, much better sound stage, better instrument separation. (both using HDAM module as output buffer)

Points of discussion:
- I left the output limiting resistor at 220K in all cases
- MAX9632 might not deliver enough current when used as single supply
- It might be better not to leave the Charge pump pins floating, but I do not know what is best (coupling to ground does not seem wise, because you'll "pollute" the ground with 1,5MHz noise.)

With nice power supply, low jitter clock and nice output buffer, the ES9023 starts to sound quite good. But all price advantages are gone, and a ES9006 or ES9016 seems like a better choice.
 
Disabled Account
Joined 2002
Hi please try 130 kOhm when using 3.6 V regs. The chip wont clip that way.

As known the ES9023 was never meant for use with symmetrical power supplies. It would have been nice if such a version would exist but I guess one has to choose a higher priced Sabre for that. Still experiments like this are nice.

Some questions:

1. How much 1.5 MHz noise do you see at the outputs when using a single power supply ?

2. Is the chip hardwired or is it on a PCB with decent layout/ground plane and decoupling ?

3. Did you compare single and dual supply configuration without the HDAMs ?

4. What happens if the symmetrical power supplies are switched on simultaneously like one would do normally ?

BTW whatever ESS Sabre DAC chip is chosen, they all need a nice power supply (or supplies which adds to the total cost with the higher ranked chips), low jitter clock and only ES9023 does not need an output buffer which makes it interesting considering it is a low cost chip too.
 
Last edited:
Hi please try 130 kOhm when using 3.6 V regs. The chip wont clip that way.

Will do, I did notice that with no resistor (as some suggested on DiyAudio) you get serious artifacts on loud music.


Some questions:

1. How much 1.5 MHz noise do you see at the outputs when using a single power supply ?

Hard to say, 1.5MHz is at the beyond limit of my PC measurement card, and on my old scope I can only see relative changes. But I see an decrease from -70dB noise in the 500-750KHz range, to -90dB when I switch to symmetrical supply.

2. Is the chip hardwired or is it on a PCB with decent layout/ground plane and decoupling ?

It's a PCB, layout is never perfect, but at least all tests were done on the same PCB. Etching didn't go that well (decided this will be my last home-made PCB :( ). Layout is verified by a professional PCB engineer. Caps are all normal ceramics, apart from the high pass filter in the output, which is MKT. 1500uF 6.3V low ESR is added to the 3.6V input (twice for symmetrical)

The 3.3V version already sounds better than the ES9023 inside the AudioWidget. Probably because I use a better XO. But this indicates the PCB is not very bad.

Schema
PCB
Photo

For 3.6V single and dual supply, the trace between the ADP151 regulator and C3 is cut. 3.6V is directly added on C3, and neg 3.6V on C2. (thus the regulator only supplies the XO.)

3. Did you compare single and dual supply configuration without the HDAMs ?
Not by listening, but noise measurements were done without HDAM.

4. What happens if the symmetrical power supplies are switched on simultaneously like one would do normally ?

With current limited supply, that works out fine, but I cannot vouch for it with a normal supply. The ES9023 has some slow start mechanisms, that might cause an initial high current flow. The ES9023 does not seem to mind that it draws 500mA for a few seconds, but that is way beyond official limits. The chip does heat up more with symmetrical, than with single supply. But after 20 hours of listening, it still works fine. Current drawn is 30mA for positive, and 20-25mA for negative, whereas the single supply version only draws 30-35mA. So there is loss. (perhaps because the chargepump goes to max when you disconnect it's 1uF cap?)

I'll do some extra measurements, because I'm not very satisfied with the low current opamp supply. Feeling bass could be improved more with a normal shunt supply. (now the version with only a 3V3 regulator has more, but floppy bass.)
 
Disabled Account
Joined 2002
OK, first thing is to change the 220 kOhm to 130 kOhm and then retry tests. You already experienced that the ES9023 clips with the wrong value and things might turn out differently when the correct value is used. It would make measurements and hearing tests easier if you would have a single supply version and a symmetrical version to compare. BTW it seems the charge pump MOSFET in the chip conducts continuously when a cap is omitted. This could explain the higher power consumption. It is clear that this is unwanted behavior and it is also operating the chip outside limits. If the chip does not burn it's OK but your experiments make it clear for me that I won't experiment with symmetrical supplies.

Comparing a single supply 3.3 V single supply version with ADP151 (excellent reg) and a symmetrical supply version with + and - 3.6 V really is comparing two entirely different setups which will always be favorable for the 3.6 V version. ES9023 sounds and performs better on 3.6 V. It would be more fair to use a 3.6 V reg on the single supply version and then compare with the symmetrical 3.6 V supply version. In fact that would be the only correct way of comparing.

Some side notes:

- The XO should have a well decoupled separate 3.3 V power supply (for the single supply version that is). You will need a bead there. See our PCB for reference if you like.

- Experiment with C2 value. 1 uF really is the minimum.

- first time that I see pin 16 used. I take it makes the LED glow up when no signal is applied ?
 
Last edited:
it seems the charge pump MOSFET in the chip conducts continuously when a cap is omitted. This could explain the higher power consumption.
Probably not much one can do about that. Perhaps frying the MOSfet?

Comparing a single supply 3.3 V single supply version with ADP151 (excellent reg) and a symmetrical supply version with + and - 3.6 V really is comparing two entirely different setups which will always be favorable for the 3.6 V version. ES9023 sounds and performs better on 3.6 V. It would be more fair to use a 3.6 V reg on the single supply version and then compare with the symmetrical 3.6 V supply version. In fact that would be the only correct way of comparing.
I have made three PCBs, one with single 3V3 regulator for both clock and ES9023, one with regulator for clock, and single 3V6 supply, and one with regulator and dual 3V6 supply. I've made only one 3V6 supply, of which the negative part is not used for the tests with single supply.

- The XO should have a well decoupled separate 3.3 V power supply (for the single supply version that is). You will need a bead there. See our PCB for reference if you like.
There's a 0 Ohm resistor in the supply line, will replace that with a small inductor when I get them. The shared regulator was to keep the costs down, which was the initial goal of this small PCB. The PCB with only the regulator for ES9023 and clock does sound worse than the PCB with regulator and single 3V6 supply. It sounds boomy, a lot of bass, but uncontrolled. and the sound stage is less).

- Experiment with C2 value. 1 uF really is the minimum.
Will do, especially for single supply version. In dual supply, the external 1500uF low-esr is already parallel to C2.

- first time that I see pin 16 used. I take it makes the LED glow up when no signal is applied ?

Yep, it works as a "no audio" led, which is quite handy some times.
 
Jean, the es9012/6/8, none of them use symmetrical supplies, the balanced output swings around a DC bias equal to the analogue supply midpoint ie. AVCC/2 ~+1.65-1.8vdc depending on your preference. its a completely +VE DAC; only +1.2 or +3.3->3.6vdc supplies on all digital, logic and analogue supply pins
 
Disabled Account
Joined 2002
OK, I see. I never bothered to check the higher ranked Sabre chips in detail. The difference between symmetrical supplies or swinging around a midpoint is negligible compared to an output stage fed by a charge pump ;)

I still think it is pretty amazing what a ES9023 can do considering board space, cost and component count. I take the charge pump for granted although at first I wanted to skip the chip (dogma:a charge pump "does not belong in audio"). The charge pump does not mess up things as much as assumed. It does its job and it does so right. The guys at ESS did a nice design.
 
Last edited:
no difference at all vs true bipolar in practice, provided you have a nice way to get rid of the DC on the output.

yes the ESS guys did a nice job alright! I havent got any hands on experience with the 9022/3, so dont have an opinion, but the 12/16/18 are all great. pity some of the guys responsible for the design left the company, so we wont be seeing anything new from them, not sure if Dustin is still there either.

That being said i'm not sure what else they could do to improve on them all that much, they already severely test the builders ability to get out of them what they are capable of.
 
Today I did some modifications and measurements on the ESS ES9023 dac.

I changed the limiting resistor to 150k (I did not have 130k). Did not do A-B testing on this resistor.

I changed the neg capacitor (C2) to 10uF, this helped alot!

I added some filtering to the feedback loop of my opamp supply, because it oscilated. Now I can use national poweropamps that deliver 150mA, should be enough for these experiments. Output is clean up to 750KHz. The faults in my opamp supply, cause incorrect results on the current measurements. Therefore, I did new measurements, now with a Data Aquisition Card (NI USB-6251).

Current consumption while playing a 1KHz sine on a 320ohm headphone at max volume.:

Single supply (chargepump 1uF present)
+3V6: 34mA

Dual supply
+3V6: 26mA
-3V6: 11.4mA

If you cut the negative 3V6, positive current increases to 30.5mA, chip will revert to normal function when the negative supply is back.

If you cut the positive supply while the negative is still on, current consumption of negative goes to max. Chip does not return to normal function upon return of positive supply, and keeps using high current on negative supply. (dangerous for the chip!)

Thus the increase in power consumption if you switch to dual supply is much less than I said in previous post. Strange behavior remains if you mangle with the negative supply. If both negative and positive are switched on at the same time, or negative is switched on later, all is fine.

The larger capacitor on the neg pin (C2) helped to lower the noise on the output. Now the single supply version has better noise floor than the dual supply one.
Noise floor of the dual supply version is roughly 90dB, compared to 100dB for single version. spectrum for dual and single supply

I still do not know what to do the CP/CN pins of the dual supply version. Both contain a combination of 1,5Mhz square wave with a sinc of ca 14MHz modulated on top of the square wave. CP has a high impedance towards the NEG pin or ground. But the CN pin has 11 Ohm impedance towards the NEG pin. Thus this high frequency square wave will still pollute the negative supply. I will try to place a 1uF or 10uF between CN and GND.



On the more subjective part, listening tests!

Connected to the HDAM output stage, I actually hear very little differences between the single and dual supply versions anymore. Both sound very good! Sound stage and body are still lacking compared to the internal CS4397 dac of the SACD1000 donor platform. But in detail and in bass the ES9023 outperforms the CS4397 (not crippled by flatcables, and a crappy PCB), which is quite spectacular.

Connected directly to my amp, the dual supply version does outperform the single supply version. Bass is tighter, and vocals have more body. Sound stage and detail are comparable (and not as good as with the HDAM).


Conclusions:
On paper, single supply is better
In practice, single supply is easier and less prone to errors
With buffer stage, dual supply and single supply perform equal
Directly connected, dual supply sounds better

worth the trouble... probably not...
 
Disabled Account
Joined 2002
Conclusions:

- On paper, single supply is better

- In practice, single supply is easier and less prone to errors

- With buffer stage, dual supply and single supply perform equal

- Directly connected, dual supply sounds better

worth the trouble... probably not...

I see your setup was faulty (as assumed earlier). As you see drawing conclusions in a too early stage leads to misconceptions. You should use 130 kOhm for optimal performance when using 3.6 V. When things are done as ESS prescribes and some extra attention (separate power supply for the XO, more capacitance for C2 as you have experienced etc.) the chip will perform very good and better than CS4397. This is known for quite a while now. There is a reason that many ES9023 designs exist and why the ODAC, Subbu DAC etc. are popular.

One thing that wonders me is the input impedance of the amplifier. If it is low that could explain a part of the difference in buffered/unbuffered configuration. One can understand that the JG buffer is a good combo with the ES9023. If I still could buy the matched FETs I would build one too.

Thanks for the experiments, it makes it clear that symmetrical supplies would be nice if the chip was constructed differently but it isn't. If you mean symmetrical supplies are worth the trouble I guess they're not. The chip itself is worth the trouble of experimenting till it performs best. You will experience that if you continue with the details.
 
Last edited:
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.