Muse 4 x TDA1543 DAC: NOS sound, or...

[wwenze]

TDA1543 isn't a new name in audio; rather, it has already been discontinued for eons, Philips Semiconductor kind of died, and NXP doesn't even store the datasheet.

This IC, along with its expensive brother TDA1541, are known for being non-oversampling (a.k.a. NOS), or multibit, DACs. Or "it's not a yucky sigma-delta DAC". But while TDA1541 is very expensive and usually found only in high-end (sic) implementations, "TDA1543" is a DIP-8 chip that is very affordable and easily implementable and can be found in many cheaper products, often with multiple chips in parallel to... I dunno, improve the performance?

(You'll find out why there are inverted commas on TDA1543 later.)

Why NOS DAC?
Many people say many different things. Detailed lengthy articles can be summoned at will by Google.

What I can remember are:

- Non-oversampling. If the input says to output a voltage of 0.5, it outputs a 0.5. As opposed to an oversampling DAC, which outputs a 1.0 for 50% of the time by turning on and off.
- Oversampling DACs require a filter to convert that 1.0 @ 50% into a 0.5, and this introduces ringing in the process. Pre-ringing in particular seems to be the fault of all things modern. NOS DACs don't require a filter since the output voltage is, just, there. Unless you count the optional LPF to smooth out the aliasing, which still doesn't cause pre-ringing.
- Sounds more natural, tube-like, yadda yadda.

Why the Muse Mini TDA1543x4 DAC?
Because it's cheap. Like seriously, it costs 200RMB on Taobao, and another 50RMB for international shipping.

There are other sigma-delta-based DACs that are even cheaper, and possibly perform even better, but I got this because it is NOS. I already have good sigma-delta DACs, this NOS DAC is for education purposes, to listen to the NOS sound that is oh so wonderful.

And also this particular model has been out in the market for pretty long and many people have heard of it / own it / have talked about it / modded it / reviewed it - there's a 65-page thread on head-fi, which only adds to the fun later.

 

[wwenze]

Stock form listening

(I have SPDIF connected to both PM6005 and Muse and with the remote I can quickly switch between PM6005 internal DAC and Muse while playing the same song. My usual setup uses PM6005 with its internal DAC.)

Does it sound very different from the PM6005's built-in CS4398 DAC? No.

Does it sound different enough to be noticeable in A-B switching? Yes.
Reliably? Don't know, that would require an ABX.

Does it sound bad enough that I would care if I didn't have an alternative? No.

Does it sound bad enough that I would change it if I have better alternatives for free? Yes. In other words, this is not a so-called giant killer.

Does it give me a strange headache like onboard sound? No.

Which is already 10 points for Gryffindor, considering I paid around $50 Singaporean dollars for this DAC. So at least it is an upgrade over onboard... I hope? I mean, I have paid more for worse shyt, although that was 8 years ago and the market wasn't as competitive or accessible.

(If you want to go even cheaper, there are products using sigma-delta DAC chips that most likely measure better too, but the reason I got the TDA1543 isn't for upgrading, but to hear the supposed NOS sound.)

Higher frequencies are audibly reduced, resulting in less openness and surround and less sparkle for instruments that require the top end. The bass... feels flappy. And overall it sounds like it has more distortion in general, especially when the music gets complex and loud, soundstage is constrained, less dynamic, less space, more dark.

Still probably an upgrade over onboard sound though. And as a cheap DAC for use in general, for your new TV that doesn't have analogue out for example. It is still decent enough to be usable. Don't know how it fares against sigma-delta DACs at the same price though.

 

[wwenze]

Stock form measurements

Well I already know what I was getting before I bought this thing, because there are others who have already measured it. I bought this with the intention of modifying it (hint in previous photo) since

1) A real TDA1543 is rated at 0.018%

2) TDA1387 is almost pin-compatible and is rated 0.004%

and

3) People on the internet have wrote that the stock resistor values are wrong and you can get better sound by replacing them with a lower value. Some even gone technical and explain with calculation of why the stock values will cause the DAC to clip. We will see whether this is correct later.

Modding attempts mean visiting the datasheet. And this was when something interesting was noted...

 

[wwenze]

Clipping, or does it?

680 ohm was said to be too high a value for the I/V resistor, swapping it to a lower value helps, some say a 400+ range, some say a 200+ range. Example quote:

As each TDA1543 contributes 2.3mA with 4 in parallel we get 9.2mA as the total current swing. Into 680R that's about 6.3V peak-peak. The compliance range is from 1.8V to VDD-1.2 which in your case gives 4.68V (because VDD is 7.68). So yes you will get clipping on digital peaks. Using 470R instead of 680R will reduce your swing to 4.3V - no clipping there but you also need to set the mid-point to precisely half-way between the compliance limits. In your case this is 4.1V which will result in swinging between 1.95 and 6.25V. You set the mid-point with the resistors connected to pin7s of the DAC chips.

However, two things that are weird about this:

1) Digital 0dBFS measurements have THD in the 0.1% to 0.2% region. If the DAC was trying to output 6.3V peak-to-peak vs the 4.7V peak-to-peak that it actually can, we're talking 25% of the waveform getting clipped. This amount of clipping would not be 0.1%, but a way higher number. (At least 10%.)

This is what 10% THD looks like:
(Image source: http://www.elab.ph/forum/index.php?topic=14726.0)

In comparison, this is what I got from the Muse at 0dBFS - There is no obvious clipping:

That squarish waveform proves that this is an NOS DAC.

2) The value of the bias resistor which sets the DC bias on the output pin. The value is 390 ohm, which, according to TDA1543 datasheet, would give 7.67V on the output pin, which... forget about extreme clipping, the DAC won't even be able to generate any sound!

Yet people are saying that the DAC can work, be it stock or modded, so what is going on?

 

[wwenze]

More on I/V

TDA1543 is a current-out DAC, so external circuitry is needed to convert the current to voltage.

A common method is an active I/V stage using an op-amp, like what is shown in the TDA1543 datasheet:

Another method is to do it passively, using a single resistor. Because Ohm's law.

In a later part I'll explain why the "Keep It Simple" method does not work well in practice. For now, let's see how Muse TDA1543x4 does it.

This is the section we're interested in:

Note: Only one output channel shown. Also, there is a 2.2nF capacitor in parallel with R10 and R11.

Rbias may also be called Rref, I don't think there is any specific naming rule in this case.

R_unknown_number are the resistors located between the two output capacitors. The exact values are not critical, they make sure the DC resistance between RCA and ground is not infinity - some amplifiers' input hate that, and you also wouldn't want static to build up. Don't think there is a name for those resistors.

How this circuit works

- Vref (fixed by DAC IC) and Rbias cause Iref to flow.
- Iout is the sum of Ibias (DC) and the AC Iout. We need to set Ibias and Vbias such that Vout can swing properly.
- Ibias equals to Iref multiplied by bias current gain (AIbias)
- And Vbias equals to Ibias multiplied by the I/V resistor.
- But at the same time, we need to select an Riv value such that the AC voltage does not swing too much and clip.
- AC voltage swing is determined by Riv multiplied by full scale current (fixed by DAC IC)

So the steps for calculation would be - determine voltage swing needed. Determine appropriate Riv value. Choose a Vbias. Determine the Iref / Rbias needed.

TDA1543 datasheet:

The datasheet did not include an example circuit for passive I/V, so I can only guess what would be the recommended values.

The below will result in Vout swinging from 1.8V to 3.8V, which are the limits when running a TDA1543 with 5V VDD.

Effect of paralleling DAC ICs

Vbias is unchanged. Output swing is multiplied by number of DACs.

Explanation:
Instead of one DAC IC forcing 2.2V into Rbias, we now have 4 DACs forcing 2.2V into Rbias. So total Iref is still unchanged, although the Iref flowing from each individual DAC IC is reduced. But this does not matter, because we're only concerned about total Iref, since it means the total Ibias is also unchanged. Which means Vbias is unchanged, even if I don't change any resistor value.

Output current swing from all DAC ICs are added together, so voltage swing is multiplied by number of DAC ICs.

So if you need more voltage swing, just parallel more DACs without the need to change anything else. Convenient.

But then again you can choose the right Riv to begin with. However changing the Riv afterwards requires changing Rbias too.

Values used in Muse 4 x TDA1543 DAC

Note that Muse used 7.6V for VDD

This is when things don't make sense:

Forget about Riv causing too much voltage swing with 4 DACs, that value of Rbias does not make sense. Nothing makes sense when Vbias is higher than VDD. Nothing will make sound too.

 

[wwenze]

Poking around

But the DAC certainly is functional. All we need is a few measurements to know what is going on.

Bias gain = 1.07x
Measured full scale output voltage = 2Vrms, or 2.828Vpp
Full scale current = 2.08mA per IC
Maximum and minimum voltages = 6.7V and 1V
VDD = 7.60V

Let's see TDA1543 datasheet again

I think you already get the point by now.

Did I get lemon from eBay where they swapped out the DAC chip?

Nope. I bought this from iVox Audio's Taobao store. Which is now the seller (or owner?) of Muse. It doesn't get more real than this; my unit doesn't even have the fake Elna capacitors which are very common in China purchases.

But more importantly, the key point is the resistor values. If they simply took out a real TDA1543 and popped in this IC, it won't even work. So they knew what this IC is and they designed the rest of the circuit for it. Even units more than 5 years back are already using the same resistor values. There are probably none of this DAC out there that uses real TDA1543.

And now the fun part when you think about it,

For years people have been modding this DAC and referring to TDA1543 datasheet for the current swing and reference current/voltage values, and concluding that the stock values are wrong and choosing new values. So far I have not seen any one write that they think the situation is weird to begin with. Or any one who measured their DAC after modding and finding out the voltages are different from expectation. Do more people own soldering irons than multimeters?

At least I finally saw one person, in 2016, who RMAA the before and after performance. He didn't notice the weirdness, but he was a follower who read that the resistor values need change and then found the lowest THD by trial and error, so this is fine. It's the pioneers who did it by calculation and didn't actually verify the result that let this joke run for 5 years.

The designer is not stupid

The supply voltage and the resistor values make sense after knowing the output voltage swing is 2Vrms. Because 2Vrms is like the de-facto level thanks to Redbook Audio. They chose the resistor value so that it will swing 2Vrms, and they had to increase VDD so it can output that range. And they chose to do that even though it will reduce THD performance (More on this later). My guess is that they don't want their DAC to have a low volume, because many people associate output voltage with performance.

 

[wwenze]

Modified form measurements

Let's reduce the Riv and see what happens then

Indeed the THD drops with lower voltage swing. This might even be better than the real TDA1543, which is rated for 0.018% in the datasheet, which we know is never achieved in real life. Note that the real TDA1543's performance is only guaranteed if you use it according to the datasheet i.e. with active I/V. The output voltage compliance (AC) of TDA1543 is only ±25mV, so anything above that (e.g. 2000mV) the THD can become significantly worse. The same holds true for this DAC IC whatever it is, but it is holding up pretty nicely even with ±474mV.

Am I angry with Muse for the switcharoo? No, I bought it already knowing it is fake after seeing the resistor values.

Is this still fake TDA1543? Yes.

 

[wwenze]

Modified form listening

I tried with 222R for a while, then now with 114R

I found that with 222R, the sound shifted from "a different type of DAC" to become much more like a typical well-measuring DAC instead. With 114R, it became almost indistinguishable from the CS4398. There is a very minute reduction in frequency response and openness, although I'm not sure if that is placebo.

I guess the lesson here is that all equipment above a certain level of measured performance sound similar, or at least sound decent and acceptable. So much for that TDA1543 NOS sound - it was distortion the whole time. Unless you consider this kind of distortion a characteristic of NOS DAC. Some people actually think that way for tubes.

I still feel that PM6005's CS4398 is still slightly better. But PM6005's DAC is no pushover - it measures less than 0.0020%. Better products include Xonar DX (0.0006%, but powered by computer) which costs $100++ and LJM CS4398 (0.0005%) which would cost $100++ to build or buy with a proper power supply and case. So $50 for the Muse TDA1543x4 which comes sonically very close to those standard setters is not too bad, albeit mods (and a high gain preamp if necessary) are needed to get to that level of performance.

Contrary to my usual fashion, I'm totally ignoring measured performance this time. Because we're trading poorer measured performance for "Ooo NOS architecture". And because onboard sound measures better but gives me headache while the Muse "TDA1543" doesn't. A pretty decent upgrade from onboard sound, a good buy especially when some entry-level DACs are no better than onboard sound.

 

[wwenze]

So why do things distort and why active I/V helps

Distortions happen because real devices are not ideal.

For current-out DACs, it's all about switching current sources. One method is shown below, but there are many ways to do it.

Source: Analog Devices. They have PDFs explaining how these all work.

A two-terminal constant current source doesn't exist in real life though. Usually they are made using current mirrors.

https://en.wikipedia.org/wiki/Current_mirror

Let's just look at the output transistor. Below is a simple current mirror using PNP transistors - so that the output pin pushes current into the load like what our DAC here is doing.

More info on operating principle in Wiki link. Basic idea is that the left branch sets the current while the right branch tries to follow that current, regardless of the value of Rload. The voltage on the pin hence depends on Rload.
(Real current mirrors are a lot more complicated - the entire left branch is replaced with a black box, and even the area above and below the output transistor are replaced with black boxes.)

A typical real transistor characteristic curve looks like below: I didn't use those textbook pictures because they don't show that the Ic vs Vce is actually a bit curvy even when in the active region.
(Thanks for the source: http://www.theautomationstore.com/transistors/)

Since the output pin is the collector pin, Vce changes with collector current. Ideally this relationship would be linear for no distortion. But it isn't. And the lower the Vce (or higher output voltage), the less linear it gets. And also the more voltage you swing, the total length is going to be less straight.

Actually the same can be said for NPN circuits / output voltage too low / voltage-out circuits like a typical audio amplifier - essentially when you swing voltage too hard and when you swing too close to the rails, you get distortion.

But why aren't voltage-out (or voltage-controlled or voltage-gain) stages affected as much then? I mean they can usually swing a few volts without any issue. Because of feedback. Take a look at those so-called no-negative-feedback amps, they usually measure pretty badly. And many of them use emitter degeneration at the transistors, which is a form of negative feedback. They just don't have global negative feedback.

It is possible to employ feedback with current output too. See feedback-assisted current mirror. So in the end, it is still about the amount of feedback (or gain, which is then used for feedback) that you can throw at it. A typical op-amp has quite a few stages between input and output pins, hence having a total open-loop gain of typically 100,000x that can be used with global feedback. (And this does not count whatever local feedback that is already being used inside.)

So end of the day, is it possible to make a current-out DAC with high linearity while having high output voltage swing? Yes. But why would you want to do that when you can achieve the same performance by using an active I/V stage, which can use as low as a few transistors to more realistically a few tens of transistors. Easily integrated into the IC.

Active I/V - easy solution for the wall of text above

So, basically the wall of text only *tried* to explain why things distort when you swing voltage too hard. And it ends off with the irritating engineering conclusion "it depends on the implementation".

But the entire issue can be avoided simply by making sure the output voltage doesn't shift regardless of output current. Enter active I/V.

According to op-amp theory, the op-amp will try to make sure the two inputs are at the same voltage.

Now, the non-inverting input is at Vref. Which means the inverting input - the DAC's output pin, will forever be at Vref. Current from the DAC momentarily changes voltage on inverting input, causing op-amp output to shift in the opposite direction which then causes voltage at inverting input to drop. At the stable state, inverting input voltage = Vref, Vout = Vref - Iout*Rfeedback