Topping D900 DAC Review

[Dobert]

You should qualify what "playing natively DSD without intermediate conversion" means. To the best of my knowledge, conversion from DSD to analogue was hardly ever made without signal processing of some sort since day one of this format. But if you means decimation of DSD to low sample rate PCM for the purpose of conversion to analogue, that is actually a very rare occurrence.

If you wonder what happens when the 10 bands parametric equalizer is used, that is an entirely different topic. If you intent to do some equalization on the audio pass-band directly on one bit data in the digital domain, you obviously end up with lengthier numbers than one bit, which is very very taxing for processing power. According to the relevant picture published on the Audiophonic webpage about the Topping D900, the parametric equalizer works only on PCM input up to 32 bits/192 kHz, if I understand it correctly:



So, no miracle indeed at this price point and silicon bill of material: no direct computing on one bit data can be performed in this DAC when the PEQ is to be used.

See my post. That R2R section is quite clearly not a single bit modulator. It looks like a multi-bit modulator just like all the chips do because single bit modulators have inherent limitations that make high THD+N practically impossible.

Okay, looked deeper. Its thermometer coding. Effectively its a 5 bit quantizer with dynamic element matching just like any chip level DAC. There is nothing really 1-bit about this other than the marketing which as we know is highly effective with audiophiles.

 

[Dobert]

Thanks Amir. I've been looking forward to this test for a number of reasons.

It's 1 bit DAC, and I hoped the measurement might show some difference with delta sigma DACs, but they're all very similar indeed.
Topping's B100 has high dynamic range if you feed it a high level line input, and I wondered if the D900 was Topping's solution to that.
Topping's own (normally trustworthy) measurements show only slightly higher noise floor for the 13V output (so they're not simply waving it through as the same as 5V).
Topping's (A-wtd) DNR measurements are read off the APx555, and don't show any DNR advantage over the 5V output.
However although the 5V maths are consistent, 20log(13/1.6E-6)=138dB, and I guess I was secretly hoping this might have been the DAC to set a DNR record.

Ah well.

There is nothing 1-bit about this DAC other than the marketing.

 

[Dobert]

The elder Delta-Sigma DACs (most consumer chips in the market) are 1bit DACs. The most modern and better chip DACs use more than one bit so a 1bit DAC is generally no advantage. Maybe in linearity but especially the D900 is not so strong in this discipline.

The special thing here is that Topping did not use a Chip DAC but built its own. By concept it's not too complicated. The upsampling can - and probably is - done in a FPGA. This is simple number crunching, some cut-off filters and an N-stage integrator for noise shaping. All can be realized in the digital domain as for example HQ-Player also does.

At the end you have a high frequency 1bit digital pulse train that can be treated with a simple lowpass filter to create the analog output. This is conceptually the same as a DSD stream coming from the SACD.

As the shaped noise gets visible above 50kHz I guess the output will be something comparable to DSD-128 or higher als in DSD-64 the noise starts getting in at a lower frequency. Depends a bit what integration filters they use for Delta-Sigma.

Except this is not a single bit DAC. It is a multi-bit modulator marketed as single bit. You can see it in the output structure. They may be all weighted the same, but the digital engine just controls how many are on to achieve multi-bit modulator like performance. That way they are not totally lying about 1-bit but really are.

 

[Dobert]

Yes, a DAC with 2^15 output voltage levels, but operating at 2x the sampling rate can output a signal with 2 ^16 possible levels. Many of the earliest CD players had 14-bit DACs with 4x upsampling to play back the 16-bit signal.

1-bit DACs take this to the logical extreme by having DACs with a single bit of precision but operating at much higher sample rates - for example 2.8 mhz for SACD delta-sigma encoding up to 22.6 mhz for DSD512 encoding.

Many of the best DACs today are 3-bit or 4-bit thermometer DACs which are operating at very high sample rates.

And so is this one, looks like 5 bits counting the parts. It absolutely is not 1 bit. In chip DACs they implement true 4-6 bit output stages. This looks like it may be 32 all around the same value (functionally equivalent). Any way you look at it, it is a multi-bit modulator. They admit it as much in their literature.

 

[Dobert]

For it's type it's exceptional good, i agree. Most 1bit dacs that cost a ton more don't even get close. It's not perfect, but that is hard to do dicrete, then you need a good IC.

It's a niche product, and in it's niche it has a vey high price/quality rating. But not what i would buy, i don't mind IC's on the right place and in a dac is definitly the right place, and it's a lot cheaper for the same performance...

They don't get close because often they are true 1-bit converters. This is not. It is evident from the circuit images and their marketing literature. It is just a take on a multi-bit modulator that they can claim is 1-bit but really is not because, you can't make a true 1-bit unit reach these levels.

 

[Roland68]

With users like you, I seriously wonder if you're a manufacturer, or why you're not interested in improving the repair and warranty conditions for these devices.
Ask yourself why you're more interested in proving that the repair and warranty situation can't be improved, just like the often far too early failures, than in helping to change that.
Or are you really afraid that the devices might become 0.5–2% more expensive as a result?

And again, since you seem to have overlooked this:
I'm not talking about a longer warranty, but about a repair option for, say, 10 years at acceptable prices for manufacturers and customers, or repair shops and customers. That's a huge difference.

We ourselves are an industrial company that manufactures high-energy products and devices at our location in Germany. We regularly service devices that are over 30 years old and still repair our products even after 30 years. Sometimes new components or circuits are used because older parts are no longer available. This is also the case for many of our industrial customers.

Since our products are used worldwide, including by capacitor manufacturers, for long-term, stress, and aging tests in the development and production of electronic components and circuits, I have a much deeper and more comprehensive understanding of this industry. This also applies to durability and potential failures.

As @Scytales already mentioned, Bryston previously offered a 20-year warranty, which is still valid for older devices. For newer devices, they now only offer a 10-year warranty, which is still very good. However, warranty claims with Bryston are just as rare as with Lake People/Violectric or RME, for example.
I know that Lake People/Violectric also repairs devices over 20 years old at reasonable prices, and they are based in Germany.
So it doesn't seem to be too difficult.

 

[boXem]

While this claims to be a 1-bit DAC, nothing about that R2R section says 1-bit. It looks a lot more like a multi-bit modulator. That would explain that high performance it achieves. I have my doubts about those THD+SNR levels with a pure 1-bit, technical prowess or not, at some point physics wins. A well designed multi-bit modulator though .. that makes total sense.

View attachment 520266

The modulator is a chip (CT7302 ), as written last week. This picture is not the modulator but the converter 1 bit -> analog.

 

[Dobert]

The modulator is a chip (CT7302 ), as written last week. This picture is not the modulator but the converter 1 bit -> analog.

The picture is the analog part of the modulator and that is not a single bit. It most definitely is not 1 bit -> analog.

The CT7320 is not a modulator, it is a translator / ASRC. To clarify it can natively transmit a DSD stream or convert PCM to DSD and do ASRC, volume, but it is not a modulator per-se. Maybe Topping calls it a modulator or some AI did, but that would be an incorrect labelling in this instantiation.

The digital portion of the modulator is that Altera FPGA. You can even see the traces running to the coax cables that run to the analog section of the modulator.

 

[boXem]

The picture is the analog part of the modulator and that is not a single bit. It most definitely is not 1 bit -> analog.

Yes it is, it is called an FIRDAC, followed by I/V converters visible on each side of the picture.

The CT7320 is not a modulator, it is a translator / ASRC. To clarify it can natively transmit a DSD stream or convert PCM to DSD and do ASRC, volume, but it is not a modulator per-se. Maybe Topping calls it a modulator or some AI did, but that would be an incorrect labelling in this instantiation.

The "translation" PCM -> DSD is done by a modulator inside the CT7302, the DSD stream (1 bit ...) is then sent to the FIRDAC.

Topping are not the first ones to do so but usually manufacturers use an FPGA or a DSP instead of the CT7302.

 

[Dobert]

Yes it is, it is called an FIRDAC, followed by I/V converters visible on each side of the picture.

The "translation" PCM -> DSD is done by a modulator inside the CT7302, the DSD stream (1 bit ...) is then sent to the FIRDAC.

Topping are not the first ones to do so but usually manufacturers use an FPGA or a DSP instead of the CT7302.

The action of converting from PCM to DSD as done in the CT7302 is NOT the multi-bit modulator that forms the actual analog portion of the DAC that is accomplished by the Altera FPGA clearly shown, resistor / analog network shown in the images. The CT7302 is the SD modulator part of the design.

This is just yet another way to do a multi-bit modulator, just like in chip DACs. They didn't miraculously find a way to eliminate all the issues of a 1-bit DAC, they just found a way to make what is really a multi-bit DAC, that they felt comfortable enough to market as a 1-bit DAC.

I should be clear, there is nothing "1-bit" about a FIR modulator, it is still mathematically a multi-bit modulator which is why they can achieve the high SINAD. No true 1-bit would on its own. They are also most definitely not simply feeding DSD into the output as again, those specs would not be met. That is why the Altera is there, to statistically spread the "bits" around through the filter. That is what their PSRM is effectively.

It is great implementation engineering, but they have not accomplished any 1-bit miracles.

 

[boXem]

You work in sales right?

 

[RichB]

The D900 clearly offends some, jeez Louise. Don't buy it then.

The volume control is relay based, that is the only part that I think could be called R2R.

- Rich

 

[Scytales]

See my post. That R2R section is quite clearly not a single bit modulator. It looks like a multi-bit modulator just like all the chips do because single bit modulators have inherent limitations that make high THD+N practically impossible.

Okay, looked deeper. Its thermometer coding. Effectively its a 5 bit quantizer with dynamic element matching just like any chip level DAC. There is nothing really 1-bit about this other than the marketing which as we know is highly effective with audiophiles.

I am not sure why you replied to me in that fashion whereas I had replied to another person about an entirely different topic that the ones you wrote in the above quoted message.

Suffice to say, after having read your other posts, that I think you may very well be over-confident in your interpretation of mere photos of the inside of the Topping D900.

What make you think the row of 32 resistors per phase you can see on the photo forms an R2R ladder? Or that the modulator used by Topping has a 5 bit quantizer? It could very well be another type of signal processing, similar to this one, which is a very well known open hardware project (Signalyst DSC-1):

The whole idea of this DAC is to average 32 successive 1 bit data. This averaging is obtained through resistive summing with 32 equal value resistors. This has an effective low pass filter effect, which is exactly what is required to demodulate the signal of interest embedded in a 1 bit bitstream.

By the way, Philips implemented the exact same idea, although by summing only 4 successive 1 bit data and in a different way than a resistive summation, in its very first SACD player, which was sold under the Marantz brandname (Philips owned Marantz at that time): the SA-1.

A variant of this idea is to use un-equal resistors to set coefficients in order to obtained what is effectively an analogue FIR low-pass filter. This technique was implemented by Burr Brown in its very first DSD DAC, the DSD1700:

On the above drawing, which is out of a patent application (about another type of DAC) written by Burr Brown design engineers, each "D" is a delay cell.

In all the above systems, the digital signal itself is always 1 bit. There is no multibit quantization. Thus, it is not misleading to call those systems "1 bit", it seems to me.

 

[danadam]

It is quite easy to demonstrate that each doubling of frequency can remove at least a bit of required sample depth.

Unless you apply some noise shaping, it removes only 0.5 bit, i.e. the noise in the original bandwidth is reduced by 3 dB, not 6. You need quadruple the sampling rate to get 1 bit.

 

[TurtlePaul]

Unless you apply some noise shaping, it removes only 0.5 bit, i.e. the noise in the original bandwidth is reduced by 3 dB, not 6. You need quadruple the sampling rate to get 1 bit.

When replying to the question (I am paraphrasing): “what use is a 1-bit DAC when my music is 16 or 24-bit” one need to moderate the level of abstraction of the reply.

 

[Steven Holt]

@JohnYang1997 could you please step in and clarify this?

 

[boXem]

that is the only part that I think could be called R2R.

Unfortunately, even this would be wrong. R2R naming comes from the arrangement of 2 resistors where one is the double value from the other. That's indeed the case in R2R DACs.
Resistors in relay attenuators have unfortunately a more complicated relationship between them.
https://en.wikipedia.org/wiki/Logarithmic_resistor_ladder

 

[taner]

They are asking too much price for a device that hasn't implemented a high quality linear power supply, TCXO clock and external 10MHz clock input. I think Gustard gives justice for their higher price devices in these departments

 

[VientoB]

They are asking too much price for a device that hasn't implemented a high quality linear power supply, TCXO clock and external 10MHz clock input. I think Gustard gives justice for their higher price devices in these departments

Why would you want an external clock input? Or a TCXO clock?

 

[SoySalsaPicante]

Why would you want an external clock input? Or a TCXO clock?

Demands of the anti-ASR establishment.