I was a little confused by some of the points in your reply. So I've tried to respond as best I can.
WAV is just file name for a raw PCM representation. I'm making assumptions as to how Sony have implemented their upsampling, I don't actually know. The DAC is listed as being able to handle 32bit float wav, but that doesn't mean that the DAC is doing the conversion (or synthesis as you put it). It might be software based or even being sent to a different chip in the Walkman (I suspect the CPU as its an ARM chip and Float processing is bread and butter for modern chips, lighting fast and very accurate
https://developer.arm.com/architectures/instruction-sets/floating-point ).
This is riddled with misinformation. The DAC is the only thing that converts a sequence of numbers into the analog signal that drives your headphones. An ARM CPU can be used to further process the numeric data (i.e before being passed to the DAC, but it is the DAC that synthesizes the waveform.
No, the spec does not state the DAC handles "32-bit float wav". The spec states the PLAYER is capable of DECODING 32-bit float (single precision) WAV files. That does not necessarily mean IEEE standard single precision values are shipped to the DAC for synthesis.
Yes, you are right though that most modern CPUs have an FPU and perform floating point computations. FPGAs for example usually do not unless the author of the RTL implements their own floating point engine (of course commercial IP may also be available for licensing).That does not mean the DAC also accepts floating point values, and often times those floating point values are quantized to fixed point integers before driving the DAC.
Also I'm confused by the use of Quantized at the output stage? Quantiztation is performed when the signal is converted from Analog to Digital.
https://en.wikipedia.org/wiki/Quantization_(signal_processing)
32 bit float is not just 8bits more;
In my post I do mention that most audio uses 32bit float vs int and therefore I
assume Sony is doing that here.
That is one form of quantization. Quantization noise is introduced at any step where precision changes. Rounding error is another type of quantization noise. Internal processing may use single precision floating point, sure... Most people model in matlab for example this way. And you're right I'm thinking from an FPGA/ASIC perspective instead of CPU - you'll have to excuse me for this, it's what I do for a living. So yes, internally there is single precision up until the output stage. At that point conversion to fixed point may be (and likely is) necessary. That introduces quantization noise... and while it may be much smaller than previously stated, it still exists and so my point of not playing the file as is still stands. By the way, anyone who has used matlab long enough has seen the limitations of both single and double precision floating point types... There are still limits to what can be represented by these types
Additionally I very much doubt that Sony are just multiplying/adding binary words against each other and therefore dealing with bit-growth. They would probably be using the C language and converting the integers to floats, then calling up samples from the audio buffer, using the Nyquist formula combined with an interpolation formula, filtering the finished result and then exporting the final result to the buffer. But when we are talking about quadrupling the sample rate to 192kHz - we are dealing with halves, quarters, eights which is were floating point numbers excel as they naturally work in Base 2.
Correct. Except for the "Nyquist formula combined with an interpolation formula" part. The filter used to implement the interpolation must have a cutoff frequency below Nyquist. Interpolation itself is performed by first upsampling (zero insertion between samples) and then filtering using a FIR. But yes, you are right this is likely being done in SW with floating point precision. That said, the above paragraphs still apply.
If thats how you like listening to music I would recommended avoiding any streaming services then and just use the DAP to play local files with Direct Source turned on. However you have to take into account that every DAC has a different design and therefore sound. Sony have their own DAC whereas Fiio, Drangonfly, Apple all use off the shelf DACs. Buying a Sony audio product very much gives you a different sound (one that I like a lot).
Like I said, I was a little confused by some of the things you put in your response, so if I've gotten the wrong end of the stick and I'm missing the point of your argument please let me know.
Why? The DAP should be able to play bit perfect from streaming sources as well as local files. What difference does it make if the file is located on an SD card or a server?
And this all somewhat amusing, because I have said all along I don't think the artifacts are audible. I'm just trying to explain the mindset of someone who is annoyed that bit perfect streaming is not an option. On principle I may agree with the sentiment, but you haven't seen me complain about the listening experience of the upsampling, have you?