Topping E50 DAC Review and Measurements: I Can't Go for That

It's the problem of these negative values in the sinc filter. The total sinc filter sums to 1. If the music happens to flip them from negative to positive, it may sum above 1 and overshoot. Thus it needs some headroom. Same overshoot problem happens with IIR filters too if it got negative numbers in its impulse response. I think if you run the math on how convolution works you'll know what I mean.

edit: I think RME or some other DSD DAC brands had a firmware update addressing this overshoot clipping issue a couple years ago.

I do all these because I can. I want to make the best DAC in the world, at least for my own taste. If there's a button on the table I'll press it and see what happens. I have some theories about why the ultrasoncis matters, but can't explain why -140dBfs things matters. It's going to be 2024 soon, but there are still plenty of werid things can't be explained by measurements in audio. If I can hear an improvement, I'll definately implement it in my own design.

Spoiler: edit

Just realize you didn't see where the 60 bits were coming from. 16 bit integer multiplied with 32 bit coefficients makes 47 bit integer. Adding them 4096 times adds another 12 bits, so it should be 59 bits in total. I was off by 1 bit.

Whatever bits you are dithering it down to, you should keep as many bits in the cache as possible before doing dithering. Otherwise quantization noise will happen, and it can protrude well above the noise floor into the audible range. That's why 32 bit DSPs has 40 bit accumulators. They'd wish to have more bits there if they could.

Damn it I kept assuming people know what I know. I don't want to explain the quantization noise here, or it would bore everyone out like my other posts.

I think you got it backwards on this one. The zero-order-hold is what made the NOS droop happening. Basically zero-order-hold is zero-stuffing convoluted with a rectangular filter. Anyway, the point of my drawings was to gave a quick and easy explaination of the aliasing frequencies. I forgot the word "aliasing" at that point.

I think you got all the points. Let me put it in another way, it's the non-linearity plus the aliasing causing the problem. If all the ultrasonic aliasing were all filtered out, non-linearity would have no signals to reflect back into the 20kHz bandwidth. Since it's easier to erase the aliasing than eliminating every single non-linearity, I vote aliasing for the first problem to solve.

It could be the same problem. The problem of E50 in around 300-400kHz is another problem of crappy analog filter stage. Your plot of E50 had linear x axis. If you change it to log x axis and do a 44.1kHz white noise stimulus on the E50, you might get similar result to GoldenOne's. I would like to, but I don't quite have the brain power tonight to put up quick jucky codes to simulate the effect. It's just my instincts talking, could be wrong. DAC204's response looks like it's passing the signal down to a 2x filter, and then another something filter.

Even if Weiss computed their own filter, they are still limited by the tap length allowed in the chip. There's a trilemma when designing filters with limited taps: how deep to cut, finish cutting at which frequency, and how droopy it is. I'd rather choose a droopy filter for my own taste, but most designs choose to leave out some aliasing.

 

Just got some simulation done. After 2xOS white noise at 44.1kHz:

The werid bridge between 23-65kHz is present because of the mediocore windowing function of this filter. It needs to be removed by the subsequent OS filters.

If the latter filter is made alright, it should be able to remove the right half of the bridge. Here's what I got with the above signal after a 4x filter:

Perhaps Weiss was using the shitty built in 4xOS filter in the ESS chip.

Note the middle fingers at 88.2kHz and 176.4kHz are caused by the dc offset (0Hz) in my white noise signal. They are aliased to such frequencies. Don't worry about it, because your music files aren't going to have any DC offset as bad as these. My white noise values are all between 0 to 1, which means entire half of the signal was DC offset.


Bonus:
After the 4xOS filter, all the left over shits above 50kHz should be filtered out by the analog LPF. I guess Topping cheaped out by going along with the datasheet output stage similar to this:

Which is a 2nd order filter, and the cut off points are rather high (268 kHz and 517 kHz). I think C10 is there for stability instead of mowing grasses. Thus they won't really cut much, and leaving the DSD noises around -70dB from 200kHz and above. It looks like Weiss did a better job filtering them down to -90dB.

 

Stock implementation of a cheap chip sounds cheap, not bang for buck. This ES9068 is cheaper than the ancient but still good CS4398 and the ES9028Q2M that Lynx and Apogee are using. It's more expensive than the controversial AK4490 and the okay PCM1795 too.

 

Mmm, good 'ol CS4398. Still one of my favourites in the Prism dacs, and those are the same dudes who make the dScope which is the other big name in analyzers next to Audio Precision. Infer what you will when the literal measurement guys aren't chasing the shiny chips with lowest numbers.

 

Prism sold off their interface, converter, and daw business to Tracktion, who own the Prism brand now. Prism the company is now Spectral Measurement and just makes analyzers and maybe does other things they are rumored to do.

 

Prism doesn't chase the chip number game, but they do bypass all the internal computation in CS4398 by feeding it Direct DSD. All the conversions are done in a FPGA they designed. They could substitute CS4398 with the newer chips, but they just got many more flashy new not really good sounding features to disable. When I feed the CS4398 in my Focusrite 2i2 2nd gen with properly OS music, the improvement is imminent, despite the delta sigma modulator still being active.

 

I don't know enough about Tracktion to know whether this would be good or bad for Prism products... but I'm assuming it's bad?

 

So I bought a string of Topping's cheaper DACs out of curiosity. OK, not only curiosity - I needed two DACs which were good-not-great (for miscellaneous PC usage) and didn't cost much.

What I bought:
DX1 (AK4493S) (DAC + HP amp)
D10s (ES9038Q2M) (DAC)
E30 II lite (AK4493S) (DAC)
E30 II (2x AK4493S) (DAC)
L30 II (notta-dak!) (HP amp)
E50 (ES9068AS) (DAC)

The headphone amp physically matches the E30II/E30II lite.

Compared to the other Topping DACs, the E50 is best. It's not better than the E30 II by a large margin - each has their strengths and weaknesses. The E50 does exhibit the Sabre sound (not clearest highs, bass lacking definition) but it's not the worst implementation I've come across, not by far. Are there better DACs for the money? Definitely! Would I buy the E50 again? Possibly - if I couldn't audition the alternative, I would choose the E50, it's good enough for its features and price that I wouldn't chance the unknown, but if I could test drive other DACs +-$40 of its MSRP? I wouldn't wind up with the E50.

I ended up gifting the DX1 to my cousin, but unfortunately, I missed the return window on the E30II lite and D10s - so they're in my drawer now lol, only used occasionally :/

I use the E30II + L30 II + headphones, and the E50 + headphones.
"Wait... the E50's outputs are only on the rear and are marked "RCA" and "BAL". How does that work?"

The E50's RCA output impedance is 20 ohms.
"It could be designed to only output low singe digit mA though!"

The chip they use is actually a headphone amp chip. Unfortunately I can't find the internal picture of the E50 which shows the chip... You'll just have to believe me - if I come across it I'll add it to the post. Basically, yes, it's OK to drive headphones with an RCA to 1/4" / 1/8" adapter from the rear of the E50. 600 ohm headphones work best, but mine aren't - you can go down to 200 ohms, but the lower their resistance, the lower the damping factor. While there's no passive crossover to get extremely messed up, the bass can still get mushy and be too loud.

 

I found it (in relation to my post ¹immediately above)... I believe the line outs (RCA, XLR) of most or all of Topping devices with 20 (SE) and 40 (BAL) ohm outputs, are driven by TPA6120A2 headphone amp chips. Although Topping's harmonic and intermodulation distortion specs are better than the TPA6120A2's datasheet, I believe this is accomplished with their "nested feedback" or whatever technique. I've looked at the specs of most of their offerings, and usually output impedance is 20/40 ohms, or 50/100. Oddlt, the D50 is 20/100 lol.

 

They're not. Topping usually has OPA1612 stages as per the DAC manufacturer datasheets. They can achieve those output impedances for line drive (by adding output resistors to isolate the output stage) but it's hardly a good way to drive headphones.

 

Yeah, I know it's not ideal to use a 20 ohm source to drive headphones, but in my case it works out all right. The headphones I'm using allow for damping factor of 15, which doesn't facilitate direct, complete control over the driver, but it's decent/sounds good to me. If I were to compare its sound from 20 ohms to being driven from the headphone amp (L30 II), itself connected to the D50, coming straight from the DAC the sound is just a slight bit warmer - otherwise differences are minimal if present at all (I didn't spend a long time comparing, but I have had the headphones for a year)

About the TP6102A - it's featured in the pictures provided by Topping -of the internals of two devices with 20 ohm output impedances- I guess it could be bogus (they do like to cover their nested feedback modules (or whatever they call them) for some reason (to the extent that a version or two had overheating issues because of it). I find it hilarious! lol anyone could open their device and disassemble the sealed modules if they were so inclined; all they're doing is preventing curious enthusiasts who don't want to risk breaking anything from looking, not competitors. Also, the vast majority of enthusiasts aren't going to be like "ooh, now I'm going to go and make myself something similar but to my exact specifications and never buy another device from Topping, or ANY audio electronics manufacturer for that matter.

Something interesting I've seen is people saying Topping only/mostly uses the 1612 - this isn't true. That is only if the internal shots they release of their devices are accurate, of course. If they are, then the 6863, 1612, and 1656 are probably present in about equal numbers. Another (the only other lol) Topping DAC I have is the E70 Velvet, and it sounds worlds apart from the E50/E30II/30IILite/D10s/DX1. It uses the new AK4499EXEQ chipset. I haven't opened it [yet, lol, always yet], but Topping's pictures show that the OP amp they chose for it was the 1656...

A lot of people seem to think Topping designs solely around the APx555... I don't think this is the case. I think it's a big part of their focus, but not solely their focus. Why? Because if you design solely around measurements, too often you're going to make a $200 device that sounds better than one of your $600 devices, a $1000 device that sounds the same as your $400 device. I would rank the sound quality of all of the Topping DACs I bought almost exactly as they were priced - perceived sound quality tracks almost perfectly with cost within their lineup - at least the portion of their lineup which I bought. Not saying their $200 DAC is worth $200, just that it sounds better than their $160 DAC and worse than the $280.

Worth mentioning so it's clear, the E70 Velvet is not in the above comparison - you get so much more per dollar with it than any of their (MANY lol) cheaper DACs which I bought. Imaging, detail, dynamics, bass, everything I didn't mention: better, alll better. I haven't spent tonnes of money on DACs before, but I have been in hifi shops and auditioned critically acclaimed hardware. My DAC 3 years ago was an Arcam rDAC - I was looking to upgrade, so I brought it to a local (now defunct) store which specialized in $2-30K hardware (speakers, amps, DACs) and slid it into an $8,000 amp connected to $25K speakers. It sounded good. Next in line was (I forget the name of it) a $7K DAC one of the sales people was enthralled with. It did sound a good bit better (unlike a couple other cheaper options which sounded different but not necessarily better (keeping that as short as possible lol)). I still remember how much better the salesman's favourite sounded compared to the rDAC, and the improvement that the E70 Velvet brings over the rDAC is about the same (amount= same, the DACs don't sound the same).
I haven't heard any of Topping's other higher end DACs, so I don't know if they're all as good as the E70 Velvet, but the E70 Velvet is very good. Whether that was an accident or engineering prowess really doesn't matter to me at this point - it sounds good! lol

edit: How hilarious would it be that Topping is doing what it's doing (making everything -115 to -125dB THD+n) to prove to the world that measurements (of a certain kind and past a certain point) are absolutely meaningless in relation to how audio devices sound.

edit2: for anyone who doesn't know what the rDAC sounded like: Tonal balance was ever so slightly warm. Detailed, but not extremely. Favourite/possibly the best part: its midrange. Its midrange is magical - I don't know how else to put it lol, especially the lower midrange into the upper-upper bass - excellent for vocal realism. Overall the sound was nice on the ears. If sound is bad/distorted or there are way too many HF reflections, my ears cry tears of scratchy fire

 

If this were the case, I don't see them much at all chagrined by how certain corners of the internet champion their approach to design as being counterproductive to the pursuit of good sound.

 

120 Ohms, if by 'back in the day' you mean the 1996 IEC standard.

 

I mean 10, for the amplifier
edit: I was misinformed! Dangit! lol

 

What I mean is, if you take everything they make under $500, it measures about the same, yet all sounds different. It's like they're trolling. I did start by saying "it would be hilarious if"