• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Review and Measurements of Teac NT-503 Networked DAC

wgscott

Active Member
Joined
Mar 1, 2016
Messages
179
Likes
37
We are here , not there , hence my question .... What are the sonic trade offs ..

I know it is really fun to play obtuse on the internet, but life is too short.

But since you are willing to generously share your wisdom and knowledge, perhaps we could start with your explanation for how upsampling differs from oversampling, both in theory and in its implementation in the Teac NT-503.

If the differences are inaudible (and they are to me, with the possible exception of upsampling to DSD, which sounded slightly worse), there is no sonic trade-off, ipso facto. The alleged advantage of upsampling to DSD and using the high-frequency cutoff filter is it minimizes pre-ringing. Since I cannot hear demonstrably measurable pre-ringing, it offers no advantage to me.
 
Last edited:

A.wayne

Active Member
Joined
Mar 21, 2016
Messages
175
Likes
20
I fail to see how I'm being obtuse, I asked if there was a sonic difference and why , thanks for finally giving me your opinion ..


regards
 

wgscott

Active Member
Joined
Mar 1, 2016
Messages
179
Likes
37
There is nothing in there that wasn't posted previously.

I still want to know why up-sampling is a "no-no." I don't wanna get no audiophile-arse-whooping.
 

DonH56

Master Contributor
Technical Expert
Forum Donor
Joined
Mar 15, 2016
Messages
7,835
Likes
16,497
Location
Monument, CO
I'm not seeing all the responses... Upsampling implies some manner of interpolation between (starting) samples, so essentially some algorithm (which can get fairly complicated) is predicting what the new samples should be. The algorithm is not always going to get it right, natch, but in practice I'd guess (do not know) differences would be inaudible unless something is really hosed.
 
OP
amirm

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
44,368
Likes
234,386
Location
Seattle Area
I know it is really fun to play obtuse on the internet, but life is too short.
I don't think Wayne is playing with you. He is asking a question. Let me answer a bit because I think you spoke past each other.

Many DACs internally run at higher speed than the sample rate of the audio. By "oversampling" this way, they get to run their core DAC at lower resolution which is easier to implement. This is because the faster you sample something, the less amount it gets to change its value. So you get away with fewer bits for the digital to audio converter. How this happens requires digging into the data sheet of the DAC silicon.

What I tested was not that. But rather, software (DSP) algorithms inside the DAC (box, not silicon) that prior to handing the PCM data to the DAC (silicon), changes its sample rate. I tested 8X upsampling and DSD upsampling. The quality and fidelity of such upsampling is determined by the hidden algorithm inside the DAC (box). So other than measurements, we can't tell what they are doing.

The idea of upsampling is both marketing and technical. Marketing one is the assumption by most everyone that bigger is better. And DSD is better than PCM. So when a DAC takes 44.1 Khz and converts it to 186 Khz, people "feel" better about it and likely read into it better sound.

The technical reason to do that is that by upsampling, you no longer need to have the sharp filter that is used for 44.1 Khz to stop aliasing. So in theory, better sound can be had -- all else being equal.

On the Teac NT-503, I briefly tested upsampling in PCM and didn't hear anything. With DSD, I thought there was some degradation but my testing was so casual that it is not worth relying on. Changing settings causes muting which makes comparisons very difficult.
 
Last edited:
OP
amirm

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
44,368
Likes
234,386
Location
Seattle Area
Ah, forgot the main point. :) That oversampling in the DAC silicon is a different process than upsampling in software. When discussing, we should keep these separate for clarity even though the core signal processing operation is similar.
 

A.wayne

Active Member
Joined
Mar 21, 2016
Messages
175
Likes
20
Ah, forgot the main point. :) That oversampling in the DAC silicon is a different process than upsampling in software. When discussing, we should keep these separate for clarity even though the core signal processing operation is similar.

Thanks Amir and separate we shall ...... :)
 

Jakob1863

Addicted to Fun and Learning
Joined
Jul 21, 2016
Messages
573
Likes
155
Location
Germany
In addition to DonH56´s post; any process that alters the sampling rate (compared to the original one) is called resampling.
Upsampling and oversampling denote processes where the generated new sampling rate is higher than needed considering the bandwidth of the sampled original signal (which by definition has to be a bandlimited signal).

Historically the term oversampling is linked to resampling processes that raise the sampling frequency by an integer factor (going back to the beginning of the CD era, when Philipps compensated the lack of a 16 bit DAC-IC with a 4 times oversampling and "noiseshaping" digital filter to avoid inferior technical data in comparison to Sony and other japanese manufacturers).

Upsampling denotes a resampling process that raises the sampling frequence by an arbitrary rational factor.

Ideally there is no new content generated during the up- and oversampling process although it is in a mathematical sense an interpolation process, but if we could meet fully the requirements of the Shannon theorem we would have the case of ideal bandlimited interpolation which means we were reconstructing the original waveform and generate new samples exactly matching this original waveform.

In reality we can´t meet Shannon´s requirements and in addition we have quantized samples; amplitude quantization is an inherently nonlinear process and we have to calculate the new samples which means we have to consider limited wordlength, limited length in time of impulse responses of real filters, truncation and dithering and even noise shaping.
 

DonH56

Master Contributor
Technical Expert
Forum Donor
Joined
Mar 15, 2016
Messages
7,835
Likes
16,497
Location
Monument, CO
My only nit is that oversampling does not in general mean resampling, at least to me. Resampling, which includes upsampling and may use oversampling, is sampling the data again, usually at a different rate, but sometimes just to resynchronize and isolate the incoming data stream. Oversampling can be just sampling the incoming (analog) signal (the first time) at a higher rate than needed to meet the Nyquist-Shannon theorem. In practice upsampling includes some form of interpolation, anything from linear to polynomial to various more complex functions (sinusoidal operations and such).

One other vexing thing in ADC (and DAC) designs is that noise within the circuits (down at the transistor level) is typically both wideband and not bandlimited. For example, to accurately acquire a 20 kHz signal, your input circuits through the actual sampling switch will likely have much more bandwidth than 20 kHz. Say it's 40 kHz (probably higher but just say) and you sample at 40 kS/s; then, the noise from 20 kHz to 40 kHz is aliased to baseband (0 - 20 kHz) and adds to whatever noise was already there. Oversampling on the surface can help that by not aliasing as much noise, but remember the actual sampling circuits need even more bandwidth even if the buffers up to the sampler do not, so there will still be some additional noise.

Finally, a reminder that by default an ADC and DAC mean something a little different to me than to the average audiophile. My context is at the device level, the acutal ADC/DAC inside the box and all the little transistors and passive elements inside the IC, not the box you plug your cables into to capture or create an analog signal. The box includes much more than just the basic ADC/DAC itself, like clock circuits, DSP/filters, and input/output buffers, though some or all of those may actually be inside the ADC chip itself.
 

mindbomb

Active Member
Joined
Nov 11, 2017
Messages
284
Likes
175
Upsampling reclocks the data, so the 8x upsampling should improve jitter performance, particularly with spdif. Usb is typically already asynchronous, so I'm not sure if that will benefit.

I think maybe the internal power supply might be hurting the performance of the teac dac. There also is a cheaper model from teac, the ud 301, which seems like it would be pretty similar performance wise, but with less frills and cost.
 
Last edited:
OP
amirm

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
44,368
Likes
234,386
Location
Seattle Area
As is common with these mods, they never provide any measurements.

I also love this statement: "The DAC is capable of turning 16/44.1kHz data from a conventional CD into 11.2MHz DSD quality playback (that's 256 times more data). "

That "256" is a ratio of 11.2 Mhz sampling to 44.1 Khz. What they don't realize is that the bit depth shrunk from 16 to 1 at the same time! So no way there is "256 times more data." Indeed there is no more data as in real information. Just a transformation from one configuration to another.
 

Dismayed

Senior Member
Joined
Jan 2, 2018
Messages
386
Likes
404
Location
Boston, MA
I just purchased the UD-301. Straight conversion without any upconversion sounds fine to me. Maybe there are technical advantages to upconversion, but I haven’t heard them yet. But I’ve only had this DAC for a few days.
 

rmo

Member
Joined
Jun 25, 2018
Messages
67
Likes
52
Ok, here is some more measurements before I rush to ship the unit back.

First, the question of USB input vs S/PDIF. Please note that my USB test file may be different because it is produced differently. Still, as you see, there is really no difference in using USB and S/PDIF when it comes to low level accuracy: (all graphs are for balanced XLR output rather than RCA posted before for completeness):

View attachment 9459

The pair of graphs by the way, are for left and right channels.

I then turned on upsampling to 8X, still using USB input. This is what I got:

View attachment 9460

Now I am showing one channel only. Again, seems to make no difference.

Now look at what happens when I upsample to DSD:

View attachment 9461

Hell broke loose! :) Low level accuracy has degraded substantially (in red). Hard to see it as a sine wave anymore. In addition, the waveform is shifted down which indicates negative DC offset (constant negative voltage).

Also, the levels are higher which in a listening test without control, may make people like DSD better even though objectively the waveform is degraded fair bit.

No time for further investigation. Have to see if my Exasound has DSD upsampling and does the same thing. Food for thought!
Hell broke loose as it should . Upsampling to DSD is a waste of time. It leads me to believe that the human ear likes listening to some distortion once in a while.
 

Paolosnz

Member
Joined
Mar 13, 2020
Messages
31
Likes
6
Hi I'm looking for information on the brother ai-503 which should have some components of the nt-503 in common. Do you have experience in this regard?
 

jkorten

Member
Joined
May 22, 2017
Messages
24
Likes
12
So I bought this NT-503 - mainly for the connectivity options that would allow my sons to plug in their laptops and bluetooth in their phones when they visit. I recently bought a 10mHz clock for this and I cannot believe the difference the external clock makes in the sound. Complete wall to wall sound, velvety, totally holographic. I don't understand why they wouldn't have built this technology in. It isn't that hard to get a low phase jitter 10mHz clock I would think. But for those of you with this unit - it is a different beast with the external clock. Holy moly.
 

ElNino

Addicted to Fun and Learning
Joined
Sep 26, 2019
Messages
557
Likes
722
So I bought this NT-503 - mainly for the connectivity options that would allow my sons to plug in their laptops and bluetooth in their phones when they visit. I recently bought a 10mHz clock for this and I cannot believe the difference the external clock makes in the sound. Complete wall to wall sound, velvety, totally holographic. I don't understand why they wouldn't have built this technology in. It isn't that hard to get a low phase jitter 10mHz clock I would think. But for those of you with this unit - it is a different beast with the external clock. Holy moly.

I'm curious how this is implemented internally. The NT-503 already has two sample clocks for the 44.1 and 48kHz families, and only a single 10MHz external clock input. Does it switch over entirely to this single external clock with a PLL frequency divider? Why didn't they go with a 14.112MHz external clock input, which would allow a clean frequency division for both 44.1 and 48kHz families?
 
Top Bottom