• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

ESS IMD hump with AKM DAC? (specifically AK4499EXEQ in Topping E70 Velvet)

mike7877

Addicted to Fun and Learning
Joined
Aug 5, 2021
Messages
739
Likes
157
Before I start talking, here is l7audiolab's IMD chart for Topping's E70 Velvet, which uses the AK4499EXEQ chipset ((((AK4499EXEQ is post-fire AKM 4499 - it is two chips now, called: AK4499EX and AK4191EQ. Digital processing (ie. oversampling and other optimizations) is done in the 4191, and digital to analog conversion and nothing else at all (no, nothing) is done in the 4499. Separation why? Mainly to reduce noise from interference from proximity of different circuit types)))

1710192459221.png


For comparison, below is l7audiolab's standard E70 performing the same test (E70 uses ESS's ES9028PRO). I used l7audiolab instead of Amir's to have both charts from the same place.
1710193086088.png


The ES9028PRO version of the E70 seems to have a tiny bit of ESS IMD hump in the -32dB to -23dB range, which disappears completely by -20dB, followed by the common and non-problematic slowly increasing distortion from -18dB to -8dB, levelling between -8dB and 0dB, at -112dB. We'd probably see the same, slow, increase in distortion on the E70 Velvet if its IMD hump didn't so drastically affect the shape of the line.

The IMD hump of the AKM chip seems to rear its head at higher level than the [typical] ESS IMD hump. The E70's ES9028PRO's little baby bump of an IMD hump is in the usual position, maybe 2-3dB lower than usual. I think the top half of the E70's ESS IMD hump was completely flattened, and the bottom half remains, so it just looks lower than usual. Anyway... this isn't about the E70, it's about the E70 Velvet, and its IMD hump!

So the E70 Velvet's hump is much more defined than we're used to seeing from ESS.

OK, that's all I have to say on the issue! Lol, I've observed it, and that's all I know!

Oh, I do know one other thing - Techpowerup, they reviewed both the E70 and E70 Velvet. In the process, they opened them both up. They didn't fully disassemble them because there were some difficult to access ribbon cables which connected stuff to the front panel, but they pulled out the boards as far as they could and looked. They said the internals were essentially the same between the devices, except for the DAC part. edit: oh, and ver. Velvet has a slightly upgraded power supply for one of its parts. Not a huge deal.

I think it's great that there are two DACs sharing the same platform - it gives a good opportunity for analysis!

So what's everyone's thoughts? I think this is a really rare find - is it not as rare as I seem to think it is?


To add: irrelevant but I'm curious if anyone knows: have non delta-sigma DACs been observed humping IMD like ESS?
 
Last edited:
I've had the velvet e70 and one reviewer said their unit had an echo sound to it. Mine sounded a little like the few ms of delay some singers put on a mic, or a DJ would use for fuller vocals. I forget which filter I used, but compared to the D10b I had right there it was different. Going to keep tabs if a firmware update is coming or other people notice it also. I'll be looking for another to see if it was just the unit I had.
 
It seems this implementation of the AKM 4499EX is also showing the IMD hump which showed up in the measurements of the SMSL D400EX.


JSmith
 
Even if the hump doesn’t show up, that doesn’t necessarily mean the hump does not exist. AP is too noisy to capture hump in some DAC. Here is the ESS Hump in D6s
 

Attachments

  • Untitled-1.jpeg
    Untitled-1.jpeg
    198.2 KB · Views: 137
IMD+N vs Level plot is *not* ideal to identify a hump-like distortion characteristic, even though it's the measurement the name came from initially.

A set of 1kHz spectra vs. level is much better and allows to look way deep down into the noise. Even better yet is to look at the distortion residual in the time domain -- sadly that requires more elaborate techniques:

The discrimination criterion for ESS hump-like distortion is that above a certain level the distortion, notably the higher harmonics, first increases but then stays pretty much constant (in absolute level). Since the fundamental signal increases, the ratio of the distortion signal vs main signal becomes lower. Only when the distortion absolute level is significantly larger than the noise level (which also is constant, normally) we can see the hump in an THD+N or IMD+N plot. At higher levels, closing in on 0dBFS, the "normal" distortion mechanisms, producing lower order harmonics, often start to dominate total distortion.

The underlying mechanism can be understood as a small ripple on the sample-in --> sample-out transfer function, see
and

AKM and other DACs also may have a hump (in THD+N plot only, not IMD+N), but from a totally different mechanism, see this post and following posts:

To add: irrelevant but I'm curious if anyone knows: have non delta-sigma DACs been observed humping IMD like ESS?
Don't know it if has been directly observed but of course for example R2R-DACs can certainly have very similar distortion patters.
 
I've had the velvet e70 and one reviewer said their unit had an echo sound to it. Mine sounded a little like the few ms of delay some singers put on a mic, or a DJ would use for fuller vocals. I forget which filter I used, but compared to the D10b I had right there it was different. Going to keep tabs if a firmware update is coming or other people notice it also. I'll be looking for another to see if it was just the unit I had.
That's quite interesting. If true, this could readily identified with measurements and would indicate some significant design error in the filter.
It not easy to see in Amir's standard set of measurements except maybe a close-up of the filter response.

The periodic strong ripple of the blue filter #5 below looks like it will have quite significant echo -- any linear-phase filter that has that periodic ripple in the passband must have pre- and post-echos in the time response (see https://www.diyaudio.com/community/attachments/preechoes-pdf.927396/ for some details). If it's a minimum phase filter, there is only a post echo which is is normally much better masked by the main signal but can still be audible with transients.
index.php
 
What's also notable is that the hump is greatly reduced (to the point it becomes insignificant) in some DACs when tested with DSD:

 
I've had the velvet e70 and one reviewer said their unit had an echo sound to it. Mine sounded a little like the few ms of delay some singers put on a mic, or a DJ would use for fuller vocals. I forget which filter I used, but compared to the D10b I had right there it was different. Going to keep tabs if a firmware update is coming or other people notice it also. I'll be looking for another to see if it was just the unit I had.

edit: I have the E70 Velvet and haven't noticed what you describe. If I were to guess, I think it might have been an issue of some units, not the result of the firmware config sent out with all units. For reference, I've listened to the E70 Velvet through the Topping LA90 Discrete as well as the E70/E70 Velvet's significant other: L70, to power my speakers - both are definitely transparent enough to pass through the effect I think you're talking about (called a vo-coder or something similar, started to be used on the vocals in music in the mid 70s - makes everyone's voice sound good lol)
 
Last edited:
IMD+N vs Level plot is *not* ideal to identify a hump-like distortion characteristic, even though it's the measurement the name came from initially.

I didn't quote your whole post, but I learned a lot from it, thanks

I have an E70 Velvet and an RME Babyface Pro (non-fs). Although the non-fs hasn't been measured here (or elsewhere afaik), it performs within ~1dB of the fs version, and its jitter isn't perfect (like the fs is).

Do you think there are any tests I could do on the E70 Velvet to look deeper into what's going on?

I can amplify its output with the L70 or L30 II (I don't know if you've seen the measurements of each of them (here and elsewhere), but they're both really great. The L30 II actually tops the 50mv chart at -100dB!
 
What's also notable is that the hump is greatly reduced (to the point it becomes insignificant) in some DACs when tested with DSD:



Whoever Tom Gibbs is loves the E70 Velvet's DSD performance

Is it possible to, say... on PC, resample PCM to DSD? Like, have the PCM upsampled and interpolated and line of best fit and then analyzed and output in DSD?

edit: forgot my reply to you: Good point/good to know about the hump essentially disappearing when DSD is he source material
 
Buncha humps! Lol. That's a nice looking DAC. I assume that's in non-oversampling mode?
There is a full review:

 
There is a full review:


It looks wonderful!
1710257786362.png
 
There is a full review:

So it doesn't do oversampling... it performs pretty well for non-oversampling! Everything looks pretty much normal except this raised noise floor on 20kHz:
1710258538125.png
 
That's quite interesting. If true, this could readily identified with measurements and would indicate some significant design error in the filter.
It not easy to see in Amir's standard set of measurements except maybe a close-up of the filter response.

The periodic strong ripple of the blue filter #5 below looks like it will have quite significant echo -- any linear-phase filter that has that periodic ripple in the passband must have pre- and post-echos in the time response (see https://www.diyaudio.com/community/attachments/preechoes-pdf.927396/ for some details). If it's a minimum phase filter, there is only a post echo which is is normally much better masked by the main signal but can still be audible with transients.

The E70 reviewed on ASR is not the Velvet (AK4499EXEQ) version, but the ESS (ES9028PRO), so those filters aren't applicable:

1710259301729.png
 
The E70 reviewed on ASR is not the Velvet (AK4499EXEQ) version, but the ESS (ES9028PRO), so those filters aren't applicable
OK, understood.
Anyway, the issue is there with almost any DAC chip, to different amounts of course.
Now we know the reason why datasheets seldom show close-up filter responses :
1710269347215.png

ESS allows for custom FIR filters so maybe the E70 ESS version is using that...
 
OK, understood.
Anyway, the issue is there with almost any DAC chip, to different amounts of course.
Now we know the reason why datasheets seldom show close-up filter responses :
View attachment 355960
ESS allows for custom FIR filters so maybe the E70 ESS version is using that...

Yeah, they're pretty ugly lol.

I dunno though, I'd think their included filters would be the best (or nearly the best) able to be implemented on the hardware - because why not use all the processing power available on the chip that they made? Better filtering can only make their product look better to everyone, too.

Unless I'm greatly misunderstanding how all this filtering stuff works (I don't know a whole lot so it's entirely possible lol)
 
Back
Top Bottom