Back to the roots: 14 bits (of TDA1541A)

One of the USB board customers asked for possibilities to connect this board to TDA1540. A TDA1540 chip uses the same (simultaneous data) protocol as TDA1541A, but it is 14 bit converter, and in that way I changed LE line, by moving its raising edge two bits to the left, so it triggers the output after 14th bit.

So, regular TDA1541A simultaneous data output from USB board is shown by the figure 1. LE (“Latch Enable”, blue curve) is aligned in a way its trailing edge matches the start, and its raising edge matches the end of 16 bit binary word (“Data”, red curve). The word that comes from PC can be also longer, say 20 or 24 bit, but only the first 16 bits will be used for conversion. Any data that comes after 16th bit, when converter output is triggered by LE raising edge, will be simply ignored.


Figure 1: USB board simultaneous data (TDA1541A) output


Moving LE raising edge for two bits to the left made it look this way.


Figure 2: USB board 14 bit (TDA1540) output


I had no TDA1540 at hand to verify this work (it was on the customer to confirm it), and the other day I was curious: will TDA1541A work with such a source too, and will it then operate as 14 bit converter?

And both answers were: “yes”.

So, as far as I understood, simultaneous data protocol is not really “right justified”. Instead, TDA1541A apparently uses LE trailing edge to start loading data into the registers, and LE raising edge to trigger the output, regardless of number of bits previously loaded into the registers.

And now, for the performance.

The first, and this may look surprising: TDA1541A distortion looked practically the same as if it was operating with all 16 bits. I used my own AYA II 2014 board for these measurements, and you can compare this 14 bit -60 dBFS (dithered sine wave) performance to that of regular AYA II, shown by the figure 5 at page 8 of AYA II manual. Not all the components are exactly the same in level, because TDA1541A chips were different series, so such small differences are expected, but there are no fundamental differences here – both graphs show the same general performance, and usual 0.8% overall figure.

In a word, even when it works with 14 bits only, TDA1541A shows its regular harmonic distortion behavior.

Noise floor, on the other hand, is somewhat higher, roughly speaking 6-12 dB, and that is what you would expect, given 2 bits difference. Mind you, wideband noise is usually considered benign, if not even welcome artifact.


Figure 3: 14 bit performance of TDA1541A, -60 dBFS, dithered sine wave


Subjective performance is also interesting. It is a bit different presentation than that of regular 16 bit TDA1541A, and at the moment it appears less precise and rich, and less polished – but also seductive on its own. Soundstage is wide and stable, though not that three-dimensional as with all 16 bits. But weight / scale appear in fact better. Background is quiet, tone is calm, and resolution is certainly not inferior to that of 16 bit either. Probably, not many would associate such a level of sonic performance with 14 bits.

Of course, if you think in terms of non-dithered signals, as long as you operate with 14 bits you can not play levels below -78 dBFS, and in that way dynamic range wold be nominally reduced. However, I doubt anyone ever does this anyhow (regardless of number available bits). And, apart from that, and even though I do not believe this is the point where we have to reconsider the terms as “resolution”, thinking about linearity that really depends on the accuracy of ladder network, and not that much on the number of bits, might indeed give new perspective on digital audio issues, at least those associated to the classic ladder (“R-2R”) converters.

And, for most of audio youngsters, talking about 16 bit converter today may look like talking about Pentium 286, let alone 14 bit. Still, back in 80s, talking about say tubes, or about any amplifier with 0.1 % distortion usually looked like talking about Intel 4004. Audio world however progressed since those days. Or, did it? Did we understand the lesson: it is not the number of zeroes after the decimal point in distortion figure that makes the unit sounding good – it is the topology in the first place. Figures can be useful, but only within the context of given topology.



Related topic:

Another story about TDA1541A: 384 kHz


4 Replies

  1. 14-09-2017

    Hello Pedja,

    very nice article on this topic as this is also what i found when using the TDA 1450 chips . I guess, as the TDA 1540 Chips are monolithic, better separation could be achieved using dual power regulation.
    I will let you know my findings once the boards have arrived.

  2. 17-09-2017

    Nice paper. Thank you Pedja.

    Hello Frank,

    did yoy try to feed TDAs 1540 with 16 bits simultaneous board ? Or only with the 14 bits tunned sim. board from Audial ? Does the first produce a sound or it is just impossible without changing LE Line start & stop as Pedja proceeds?

    Also this time on the reccording side : what is giving 24 bits recordings applied to TDA 1541A in simultaneous mode ?

    My AYA through an other simultaneous board can play 24 bits reccordings but I was not able to get a same rccording with both 16 bits and 24 bits alternativ (although there are reccording with 16 and 24 bits but all the one I found are each time remixed so not valid for a proper 16 bits vs 24 bits benchmark played by a 16 bit TDA1541A… not talking about sampling frequancy which is moving too : 44 vs 192 K Hz )

    Second part of the question may be off topic, but for te moment all the 24/192 playback through the AYA II (with tweaked digital front end) are very good through the AYA II, while I’m not able to say why the 8 bits in plus from the 24 bits is a limitration for a 16 bits dac chip and if the NOS vs 192 sampling is an imrpovment (i will say I may prefer the 192 sampling reccordings while not knowing if it’s because the material is remixed vs a old 16/44 ripped CD…)

    I haven’t tried yet with my TDAs 1540 !



  3. 21-09-2017

    Hello Frank,

    Yes, separate supply might be the advantage, even though we can also use two TDA1541A (and get balanced operation as well). There are also certain other differences between TDA15410 and TDA1541A, and I will have to check what we can still read about them in papers from that time.


    You can send either 16 or 24 bits word to TDA1541A, but it will always convert only the first 16 bits. When you play 24 bits recordings, it is on you to choose if you want to play it as such, and let TDA1541A truncate the bottom 8 bits, or to possibly perform dither, and convert from 24 to 16 bits. I personally do not use dither.

    In my view, sampling frequency was way important limit of original 16/44.1 (CD) format, which remained music industry through all these decades, than the number of bits.

    BTW, as of yesterday, Audial USB interface can work up to 384 kHz with TDA1541(A). Unfortunately I don’t have much recordings to use this feature, and actually listen to the music in this format, but my first impressions are good.


  4. 05-10-2017

    Hello Eldam,
    As far as i know we can not offer 16 bit words to a TDA1540.
    @Pedja, yes you are correct, we can offer 2x left cannel informatiin to one TDA1541 and make it mono ballanced even.
    Something i will try as well, as the rest of my system is ballanced anyway.
    Once done i will share the outcome.
    Happy listening!

Leave a Reply