Topping DX5 II

[Roland68]

I think Topping will eventually come to DAC with 2.1 outputs. But they can sell many other models before they do that. One small upgrade with each new model. And many will feel the need to get the new model.
They know exactly how this game works.

It's not as easy as you might think if you want it to work well.
There are now many devices that have integrated 2.1, but hardly any of them work properly. Good implementations only start at four-digit prices, or in inexpensive AVRs or devices like the miniDSP or t.racks from Thomann.

Analog is far too complex. Take a look at the ZinAmp Active Crossover with Phase Correction, for example. It would require two and would have to be completely set up and adjusted. Many people probably fail because of this.

Digital makes much more sense.
But it would only make sense before the D/A conversion. That means a DSP that outputs at least 3-4 channels, is of sufficient quality and powerful enough (computing power). This would then require 3-4 output channels or multiple DAC chips for the DAC.
And this would also have to deliver good measurement values.
Added to this are significantly larger enclosures, just for the connections and additional circuits. A reliable volume control also needs to be integrated.
And the whole thing has to be developed, which isn't exactly cheap.

 

[RHO]

I'm not saying it is easy.
They could wait with a new product until they get one with all the hot features on board, developed in a reliable way. That would take much more time between product versions.
But they don't and choose to implement small incremental improvements with every successive product release. (I certainly don't blame them.)

 

[ziggurcat]

@audiofun here is the reply to the feature request you asked me to submit in this post:
Post in thread 'Topping DX5 II'
https://audiosciencereview.com/forum/index.php?threads/topping-dx5-ii.60996/post-2347885

“Dear Sir

Thank you for your message and for sharing your feature request. We will forward your suggestion to our R&D team as a reference for future hardware and software updates.
Thank you again for your support and valuable feedback!


Best Regards,

service
[email protected]

This is the same response they gave me when I asked them about PC drivers for ARM-based Windows systems (the current drivers are incompatible, and can't be installed).

 

[czAudiophile]

Extract the file before copying to folder. Did not work for me when I went straight from the zip file.

thank you that somehow worked like a charm. Can someone answer then how to edit eq? bass preset looks awesome but need some tweaks to my taste.

 

[ziggurcat]

thank you that somehow worked like a charm. Can someone answer then how to edit eq? bass preset looks awesome but need some tweaks to my taste.

I would hold off on any PEQ until Topping releases the next FW update. PEQ settings are not saving right now.

 

[wilhelm2015]

i am not understanding this whole DC fuss. can't you just add a steep peak filter to attenuate the low frequencies and solve the DC in file problem? most people cant even hear 40hz. the device supports EQ and even if not you can add it on the computer side.

 

[ngs428]

Thanks! One more thing (sorry too many things). Would you please ask them if dx5ii has the same quantization bug as D50III, and if yes, are there plans to fix it?
I believe many people care about that as well.

Here is their reply..

“Hi,

This is a normal phenomenon and not a bug. It is caused by limited calculation bit depth, which can introduce some noise at low frequencies. Please don't worry — you can use it with confidence!


Best Regards,

service
[email protected]

 

[ziggurcat]

Here is their reply..

“Hi,

This is a normal phenomenon and not a bug. It is caused by limited calculation bit depth, which can introduce some noise at low frequencies. Please don't worry — you can use it with confidence!


Best Regards,

service
[email protected]

They're responding. That's good. Looking forward to seeing their responses to the issues you've put forth to them.

 

[Dandelion]

thank you that somehow worked like a charm. Can someone answer then how to edit eq? bass preset looks awesome but need some tweaks to my taste.

Here is their reply..

“Hi,

This is a normal phenomenon and not a bug. It is caused by limited calculation bit depth, which can introduce some noise at low frequencies. Please don't worry — you can use it with confidence!


Best Regards,

service
[email protected]

It's exactly what I was talking about some replies earlier, quantization noise is not a bug it's by design, there's a limit on how much these chips can pull, now... Arguably unless extreme gain settings is fairly inaudible, but it does reduce the amount of clean sound runway we have at disposition, it depends on what people do with their setup, if for whatever reason they boost that signal outside of the dac by say 20 db to feed speakers or whatever.... well then you'll hear it. But even then, the algorithm quality is pretty primitive. My advice stays the same: find a way to have a proper software eq working from the source, if you guys can use a full operating system then pro-q by fabfilter or tonebooster equalizer pro (also available in limited form on ios as AUv3 in case someone routes audio from a mobile daw) or kirchhoff eq, tdr slick eq M (mastering version but there's a free limited version), Melda production perhaps has a free good one... I understand the limit of this for audiophile setups, I'm open to give suggestions based on your own setup to ensure you guys get a nice eq in your setup.

I guess the way to make this modular across products and to accomplish good quality eq is to add a sharc processor, but that's easier said than done, as first thing they aren't cheap, and then they need quite a good amount of heat dissipation, on my uad with 4 sharc processors you can cook eggs in the summer, then you need to integrate it in the design of the board etc.

 

[Dandelion]

Just as thought experiment, would it be possible to have a product like the dx5 II run solely on usb-c power draw? Like, I get it, it's a limit on compatibility and whatever, but same specs, same quality, even with "just" the amplification power of the previous generation. What's the limit and how fast would they dry up the battery of mobile devices like laptops (say devices that have batteries between 45wh to 95wh)? I would really like to see something like that in the market, all I see is products in the class of dx1, smsl c100, which personally I find not quite satisfactory.

 

[Sokel]

I guess the way to make this modular across products and to accomplish good quality eq is to add a sharc processor,

About sharc processors:

M

'DSP Measurements and Rising Noise Floor'

Feb 25, 2023

In several threads related to DSP platforms I've seen requests for measurements with filters engaged. I thought it unlikely that implementing filters in a DSP would have measurable effect on noise / distortion other than the obvious change in the fundamental, however @Sokel recently mentioned that he returned his miniDSP Flex after observing noise / distortion issues with filters applied. This piqued my interest as a long time DSP user and I decided to explore it for myself.

The first issue is what filters to use and what tests to run. I decided to use filters for my LXmini + sub...

• mdsimon2
• Replies: 200
• Forum: General Audio Discussions

• 

No free meal there either, and that's not the only thing to consider.
Add volume control, etc and you can see the picture.

My advice? EQ upstream, with a PC, laptop, etc.

 

[jodrim]

I just got this device but it doesn't seem to save any settings between restarts like colors, gain, etc. I also have to select language each time it starts. Is this normal? I've upgraded to the latest firmware 1.39

 

[Dandelion]

About sharc processors:

M

'DSP Measurements and Rising Noise Floor'

Feb 25, 2023

In several threads related to DSP platforms I've seen requests for measurements with filters engaged. I thought it unlikely that implementing filters in a DSP would have measurable effect on noise / distortion other than the obvious change in the fundamental, however @Sokel recently mentioned that he returned his miniDSP Flex after observing noise / distortion issues with filters applied. This piqued my interest as a long time DSP user and I decided to explore it for myself.

The first issue is what filters to use and what tests to run. I decided to use filters for my LXmini + sub...

• mdsimon2
• Replies: 200
• Forum: General Audio Discussions

• 

No free meal there either, and that's not the only thing to consider.
Add volume control, etc and you can see the picture.

My advice? EQ upstream, with a PC, laptop, etc.

Of course, software eq will always be cleaner, that's why I said it's easy in theory but in practice the implementation can be a nightmare to do right. There's nuances though, implementation of on board sharc processors can be done better. One can't generalize by observing only 1 sample/example. Also keep in mind, if they used floating point, then the performance of this eq (I'm just speculating, I haven't seen the data of the topping eq) isn't much different in thd compared to that minisharc processor of the post you shared, I'm too lazy to modify again my code and test it just to revert it back but I believe to remember when I coded my parametric eq in floating point it was around -70db in thd while switching to double precision the quantization noise felt down to -150db which is pretty much the limit of computers CPUs, also it depends on the bit depth of the ESS chips.

 

[Sokel]

Of course, software eq will always be cleaner, that's why I said it's easy in theory but in practice the implementation can be a nightmare to do right. There's nuances though, implementation of on board sharc processors can be done better. One can't generalize by observing only 1 sample/example. Also keep in mind, if they used floating point, then the performance of this eq (I'm just speculating, I haven't seen the data of the topping eq) isn't much different in thd compared to that minisharc processor of the post you shared, I'm too lazy to modify again my code and test it just to revert it back but I believe to remember when I coded my parametric eq in floating point it was around -70db in thd while switching to double precision the quantization noise felt down to -150db which is pretty much the limit of computers CPUs, also it depends on the bit depth of the ESS chips.

It's not one sample, it's lots of them at different configurations.

Of the 12 tested and after the patrial fix miniDSP applied to their own we can see 20-30dB penalties at noise to all but 3 of them when filtering under 100Hz.
And those are all at the digital domain, these results will go downstream and the analog sins will be added to them.

Yes, implementation is everything, despite the hardware.

 

[audiofun]

It's exactly what I was talking about some replies earlier, quantization noise is not a bug it's by design, there's a limit on how much these chips can pull, now... Arguably unless extreme gain settings is fairly inaudible, but it does reduce the amount of clean sound runway we have at disposition, it depends on what people do with their setup, if for whatever reason they boost that signal outside of the dac by say 20 db to feed speakers or whatever.... well then you'll hear it. But even then, the algorithm quality is pretty primitive. My advice stays the same: find a way to have a proper software eq working from the source, if you guys can use a full operating system then pro-q by fabfilter or tonebooster equalizer pro (also available in limited form on ios as AUv3 in case someone routes audio from a mobile daw) or kirchhoff eq, tdr slick eq M (mastering version but there's a free limited version), Melda production perhaps has a free good one... I understand the limit of this for audiophile setups, I'm open to give suggestions based on your own setup to ensure you guys get a nice eq in your setup.

I guess the way to make this modular across products and to accomplish good quality eq is to add a sharc processor, but that's easier said than done, as first thing they aren't cheap, and then they need quite a good amount of heat dissipation, on my uad with 4 sharc processors you can cook eggs in the summer, then you need to integrate it in the design of the board etc.

Topping use same chip (XU316) as JDS Elements IV and the latter can do more and sophisticated EQs while preserving high precision, and JDS mentioned they are no where near the limit of the chip (they only used 4 cores of the 12).

Again, modern processors are very, very powerful, even the USB interfacing chip Topping and JDS use is powerful enough to run 30-40 bands EQ without losing bits. Again, it's already 2025, to do this simple thing, no sharc processor is needed.

It's an easy to fix bug---you just cast audio input signal's 24 or 32 bit number to 48 or 64 bits, then do the computation in that space. It's very simple.

So the chip is not the limitation here. Topping's firmware engineering is.

BTW, I don't even believe topping do their computation in 32bits, as the best DACs today can only have SNR equal to ~24 bits, and the rising quantization noise is already surpassing that level. Very likely they are only working on 24bits numbers.

About sharc processors:

M

'DSP Measurements and Rising Noise Floor'

Feb 25, 2023

In several threads related to DSP platforms I've seen requests for measurements with filters engaged. I thought it unlikely that implementing filters in a DSP would have measurable effect on noise / distortion other than the obvious change in the fundamental, however @Sokel recently mentioned that he returned his miniDSP Flex after observing noise / distortion issues with filters applied. This piqued my interest as a long time DSP user and I decided to explore it for myself.

The first issue is what filters to use and what tests to run. I decided to use filters for my LXmini + sub...

• mdsimon2
• Replies: 200
• Forum: General Audio Discussions

• 

No free meal there either, and that's not the only thing to consider.
Add volume control, etc and you can see the picture.

My advice? EQ upstream, with a PC, laptop, etc.

If I have to use EQ on PC then I have no reason to purchase the device in the first place. CD player, TV and game console also cannot route through my PC.

 

[Sokel]

BTW, I don't even believe topping do their computation in 32bits, as the best DACs today can only have SNR equal to ~24 bits, and the rising quantization noise is already surpassing that level. Very likely they are only working on 24bits numbers.

Don't confuse the analog performance to this.
Filtering is done digital, if done right you can see that they can reach 24-bit at the table of the post I linked.

To do it right though, you need a 64-bit processor, lots of computational power, etc.
Specially when done under 100Hz (the lower, the worst, sadly, as that's the region for RC) .
Available taps to each filter is important too.

The needs are also greater as sampling go up.

 

[audiofun]

Don't confuse the analog performance to this.
Filtering is done digital, if done right you can see that they can reach 24-bit at the table of the post I linked.

To do it right though, you need a 64-bit processor, lots of computational power, etc.
Specially when done under 100Hz (the lower, the worst, sadly, as that's the region for RC) .
Available taps to each filter is important too.

The needs are also greater as sampling go up.

View attachment 463168

i think you have a lot of misunderstandings here.

toppings quantization noise can be tested via analog output, which has lower than 24 bits SNR. This means their digital domain noise is worse than 24 bits. Digital noise the dominant part and significant enough to impact analog domain noise. If they are doing 32 bits calculations, digital noise can hardly reach that level.

Also data manipulation width is entirely different from address space width. A 16 bit or 32 bit processor has no problem doing 128 or even 1024 bits calculations. Neither JDS nor RME has 48 or 64 bits processors in their device. But RME’s computation is entirely in 48 bit, so you can get clean DSPed digital signal from their digital output.

Also Biquad EQ ( which topping, JDS and RME are using) is very cheap, cheap enough that it costs almost nothing. It only has less than a dozen lines of instructions. Moving data from chip to chip, for instance, costs more than EQ computation.


Again. I believe you are not knowledgeable enough to understand how powerful current Xmos chip is.

 

[Joachim Herbert]

Glad I use Roon and have no use for EQ on the DX5-II. I believe it is a very impressive device, even considering the limitations.

 

[Sokel]

Again. I believe you are not knowledgeable enough to understand how powerful current Xmos chip is.

I'm probably not, you're right.

Have a nice day!

 

[Dandelion]

i think you have a lot of misunderstandings here.

toppings quantization noise can be tested via analog output, which has lower than 24 bits SNR. This means their digital domain noise is worse than 24 bits. Digital noise the dominant part and significant enough to impact analog domain noise. If they are doing 32 bits calculations, digital noise can hardly reach that level.

Also data manipulation width is entirely different from address space width. A 16 bit or 32 bit processor has no problem doing 128 or even 1024 bits calculations. Neither JDS nor RME has 48 or 64 bits processors in their device. But RME’s computation is entirely in 48 bit, so you can get clean DSPed digital signal from their digital output.

Also Biquad EQ ( which topping, JDS and RME are using) is very cheap, cheap enough that it costs almost nothing. It only has less than a dozen lines of instructions. Moving data from chip to chip, for instance, costs more than EQ computation.


Again. I believe you are not knowledgeable enough to understand how powerful current Xmos chip is.

Single-precision (32-bit, float)
Max value:
≈ 3.4028235 × 10³⁸


Double-precision (64-bit, double)
Max value:
≈ 1.7976931348623157 × 10³⁰⁸


If you’re in 32-bit world, you can indeed get much higher noise way above 24 bit level.
Why? You are looking at approximately 7 decimal digits vs 15-17, Why you need them?
A parametric EQ’s biquad filter uses feedback (Infinite Impulse Response). This means errors in each sample accumulate over time. In float the smaller mantissa means you get truncation and rounding errors in state variables (x[n-1], y[n-1]) and these errors compound because they feed back into the output, increasing noise and distortion, especially under processing lik cascaded bands.

In double the larger mantissa reduces these accumulations by a boatload.


I give you a practical example: apple silicon M4 max (arm64), 8 bands parametric eq + low and high cut. Floating point needs irrelevant processing power and gets a -70db thd quantization noise. Double precision consumes 6% cpu and has -150db thd. I’m certain the xmos is a great (32 bit) processor, but do we agree no matter how good, we’re in front of technical limitations and magnitudes of power away?