Denon avr-x3900h preamp measurements

[alex.schroedsen]

Greetings everyone, I am excited to present a basic set of measurements for the pre-out front left and right outputs on the Denon avr-x3900h. Ever since I made the mistake of buying a PCM5102-bearing Denon avr-x3700h a few years back; I have been on the lookout for the next Denon receiver that contains a better DAC chip. I was so excited when I saw the news of a new 8-channel Cirrus Logic DAC being employed in the design of the Denon avr-x3900h. I bought one to play with and test it myself. Here are a few stats for your enjoyment until Amir no doubt releases a comprehensive review.

The conditions of the test are set up as follows

Power delivery:

Toroid ISB-1462 isolation transformer to supply power for Denon x3900h and Topping P50

Topping P50 5-volt linear regulated power supply to power E1DA Cosmos ADCiso

Laptop computer Dell Latitude 3340 running from battery

Cables:

Monoprice Premium XLR to RCA P/N: 4777

Monoprice Premium HDMI P/N 3992

Unshielded generic IEC320 C13 to C14 cables

Generic USB-C to USB-A cables

Audio Hardware:

Denon avr-x3900h

E1DA Cosmos ADCiso

Audio Settings Changed in Denon avr-x3900h:

Preamplifier mode engaged, no amp-out

Front speaker crossover set to FULL RANGE

Software used:

Rightmark Audio Analyzer 6.4.5

FlexASIO + FlexASIO GUI (I know the auto-convert setting is counterproductive, but for whatever reason setting exclusive mode for both devices yielded inconsistent results strobing rapidly between 50 dB and 110 dB noise floor. I made sure to change the sampling rate of the playback and recording device to match the tested sampling rate, all enhancements were disabled and volume was 100 percent.)

This is a very bare bones test and there are probably some issues with my Monoprice XLR to RCA cables based on the stereo crosstalk measurements and IMD measurements.

Some of the statistic recordings are showing some results that I find exciting and am thrilled to share with you all. I’m interested to hear your impressions.

I have attached the Rightmark Audio Analyzer SAV results files.

RightMark Audio Analyzer test report​

​

Testing device	Denon x3900h preamp mode full range xover 2441 r2
Sampling mode	24-bit, 44 kHz
Interface
Testing chain	External loopback (line-out - line-in)
RMAA Version	6.4.5
20 Hz - 20 kHz filter	ON
Normalize amplitude	ON
Level change	-12.0 dB / -11.9 dB
Mono mode	OFF
Calibration singal, Hz	1000
Polarity	inverted/inverted

​

Summary​

​

Frequency response (from 40 Hz to 15 kHz), dB	+0.04, -0.23​	Very good​
Noise level, dB (A)	-113.8​	Excellent​
Dynamic range, dB (A)	84.8​	Good​
THD, %	0.00084​	Excellent​
THD + Noise, dB (A)	-51.7​	Poor​
IMD + Noise, %	0.397​	Average​
Stereo crosstalk, dB	-42.5​	Poor​
IMD at 10 kHz, %	0.00172​	Excellent​
General performance	​	Good​

​

Frequency response​

​

	Left​	Right​
From 20 Hz to 20 kHz, dB	-8.01, +0.04​	-7.95, +0.10​
From 40 Hz to 15 kHz, dB	-0.23, +0.04​	-0.16, +0.10​

​

Noise level​

​

	Left​	Right​
RMS power, dB	-114.8​	-111.3​
RMS power (A-weighted), dB	-115.7​	-111.8​
Peak level, dB FS	-61.4​	-61.3​
DC offset, %	-0.0​	-0.0​

​

Dynamic range​

​

	Left​	Right​
Dynamic range, dB	+85.7​	+85.6​
Dynamic range (A-weighted), dB	+84.8​	+84.7​
DC offset, %	-0.01​	-0.01​

​

THD + Noise (at -3 dB FS)​

​

	Left​	Right​
THD, %	0.00087​	0.00082​
THD + Noise, %	0.23546​	0.23546​
THD + Noise (A-weighted), %	0.26048​	0.26048​

​

Intermodulation distortion​

​

	Left​	Right​
IMD + Noise, %	0.39748​	0.39747​
IMD + Noise (A-weighted), %	0.10159​	0.10157​

​

Stereo crosstalk​

​

	Left​	Right​
Crosstalk at 100 Hz, dB	-69​	-7​
Crosstalk at 1000 Hz, dB	-76​	-7​
Crosstalk at 10000 Hz, dB	-76​	-7​

​

IMD (swept tones)​

​

	Left​	Right​
IMD + Noise at 5000 Hz,	0.00117​	0.00126​
IMD + Noise at 10000 Hz,	0.00072​	0.00096​
IMD + Noise at 15000 Hz,	0.00268​	0.00352​

​

RightMark Audio Analyzer test report​

​

Testing device	Denon x3900h preamp mode full range xover 2448
Sampling mode	24-bit, 48 kHz
Interface
Testing chain	External loopback (line-out - line-in)
RMAA Version	6.4.5
20 Hz - 20 kHz filter	ON
Normalize amplitude	ON
Level change	-11.9 dB / -11.9 dB
Mono mode	OFF
Calibration singal, Hz	1000
Polarity	inverted/inverted

​

Summary​

​

Frequency response (from 40 Hz to 15 kHz), dB	+0.04, -0.22​	Very good​
Noise level, dB (A)	-114.4​	Excellent​
Dynamic range, dB (A)	114.0​	Excellent​
THD, %	0.00085​	Excellent​
THD + Noise, dB (A)	-84.7​	Good​
IMD + Noise, %	0.014​	Very good​
Stereo crosstalk, dB	-39.5​	Very poor​
IMD at 10 kHz, %	0.00104​	Excellent​
General performance	​	Very good​

​

Frequency response​

​

	Left​	Right​
From 20 Hz to 20 kHz, dB	-0.85, +0.04​	-0.78, +0.10​
From 40 Hz to 15 kHz, dB	-0.22, +0.04​	-0.16, +0.10​

​

Noise level​

​

	Left​	Right​
RMS power, dB	-115.1​	-111.8​
RMS power (A-weighted), dB	-116.1​	-112.6​
Peak level, dB FS	-61.6​	-61.5​
DC offset, %	-0.0​	-0.0​

​

Dynamic range​

​

	Left​	Right​
Dynamic range, dB	+114.8​	+111.7​
Dynamic range (A-weighted), dB	+115.6​	+112.4​
DC offset, %	-0.01​	-0.01​

​

THD + Noise (at -3 dB FS)​

​

	Left​	Right​
THD, %	0.00088​	0.00082​
THD + Noise, %	0.00524​	0.00525​
THD + Noise (A-weighted), %	0.00582​	0.00582​

​

Intermodulation distortion​

​

	Left​	Right​
IMD + Noise, %	0.01448​	0.01449​
IMD + Noise (A-weighted), %	0.00379​	0.00382​

​

Stereo crosstalk​

​

	Left​	Right​
Crosstalk at 100 Hz, dB	-63​	-7​
Crosstalk at 1000 Hz, dB	-70​	-7​
Crosstalk at 10000 Hz, dB	-70​	-7​

​

IMD (swept tones)​

​

	Left​	Right​
IMD + Noise at 5000 Hz,	0.00117​	0.00126​
IMD + Noise at 10000 Hz,	0.00072​	0.00095​
IMD + Noise at 15000 Hz,	0.00095​	0.00119​

​

 

[GD Fan]

FYI the test report section is just about impossible to read if the forum is set to dark mode.

 

[alex.schroedsen]

FYI the test report section is just about impossible to read if the forum is set to dark mode.

Thanks for the heads up. I'll try to find a solution tomorrow.

 

[NTK]

5+ years for your first post?

Do you know why there is so much spectral spreading in your tests? Have you tried not using FFT window?

Here is the example in the test guide for comparison.

 

[Marc v E]

Thank you for this review.

I see some big improvements in sinad compared to the previous generation.

Does any of the measurements indicate that there is a perveivable audible difference between the x3900h and the previous generation ?

 

[alex.schroedsen]

5+ years for your first post?

Do you know why there is so much spectral spreading in your tests? Have you tried not using FFT window?

Here is the example in the test guide for comparison.
View attachment 536703

Thanks for the education on this subject. I'm learning and I'll use this info to improve this post when I have some time to put the setup back together/fiddle in Rightmark again.

 

[alex.schroedsen]

Thank you for this review.

I see some big improvements in sinad compared to the previous generation.

Does any of the measurements indicate that there is a perveivable audible difference between the x3900h and the previous generation ?

You're very welcome! Some of the results are quite exciting to me. Some data points are very unimpressive such as the stereo crosstalk. I'm guessing my XLR to RCA cables may be ground to ground instead of ground to negative (in the differential pair) and it's taking its toll.

I am not sure that there will be a huge audible difference between -85dB THD+N versus -101dB THD+N. I am however quite excited to find out. I'm working on a 16 channel audio volume control project and am excited to hear what the difference will be. The new solution will be based on the PGA2310 chip where my previous volume controller has been the Emotiva UMC-1 (8 main channels + 2 zone 2 channels), which seems to have a threshold above the noise floor where content is intentionally dropped off. I can't wait to experience purely resister based volume control once again. I'll post updates here as I do some listening. No listening tests have happened with my x3900h just yet.

 

[Ruediger]

Thanks for putting in some work!

I see some interesting numbers (clear DAC progress?), but putting it all in perspective is beyond my knowledge.

 

[peng]

Thanks for the heads up. I'll try to find a solution tomorrow.

Thank you for the tests, my question is, THD seems good but SINAD (THD+Noise) is no better, in fact a little worse than the X3800H, wonder if your set up somehow introduce unusually high noise? Based on Denon's track record on the X3000H series, excessive noise had not been an issue, so I just wonder why this time.. It the X3900H really has such noise issue, that's too bad, what a waste of the much better DAC IC itself. I did noticed that you did it at input of -3 dBFS vs Amir's 0 dBFS but they cannot explain the differences.

 

[Oddball]

Many thanks for the review and looks like THD has improved on 3900H compared to 3800H. Not reading SINAD clearly though.

Not sure if you have the ability to test the new feature called Channel expander for Atmos mode, but that would also be interesting to understand if possible. Description in the manual is a bit cryptic, but in the best case it could possibly expand fixed coded Atmos signal like 7.1.2 to 7.1.4 or perhaps 5.1.6. Many thanks in advance.

 

[peng]

Many thanks for the review and looks like THD has improved on 3900H compared to 3800H. Not reading SINAD clearly though.

We actually did not have (unless I missed something) the THD figure for the X3800H as Amir measured SINAD, people rarely measure THD any more.

 

[peng]

Thanks for putting in some work!

I see some interesting numbers (clear DAC progress?), but putting it all in perspective is beyond my knowledge.

Yeah, I couldn't figure out how to the interpret his THD vs THD+Noise figures and graphs. They don't seem to correlate like what we typically see.

 

[Oddball]

We actually did not have (unless I missed something) the THD figure for the X3800H as Amir measured SINAD, people rarely measure THD any more.

Ups got caught - just unscientific assumption. I think lots of us expect better SINAD though. My best guess is it will be a 1-2dB below 100dB. Very good but not upsetting the north of 100dB upper tier models.

 

[NTK]

I am pretty sure the preculiar looking spectral (FFT magnitude) plots as mentioned in my post #4 were caused by mis-configured FFT analysis settings.

The scalloping in the THD spectrum plot looks like it was due to the analysis signal having zeros padded to it. Below are FFT magnitude plots generated with and without zero padding. Using more samples (i.e. increasing the FFT length) should also help with increasing the frequency resolution and thereby visually reduce the amount of spectral spreading (spectral leakage) in the plots.

Spoiler: Julia code used to create the above plots

# Reference ASR post: https://www.audiosciencereview.com/forum/index.php?threads/denon-avr-x3900h-preamp-measurements.71682/
using Plots
using FFTW
using DSP

dB(x) = 20.0log10(abs(x) + 5.0eps())

fs = 384000.0
nsamples = 2^16
f_signal = 1000.0

t = collect(0:nsamples-1)/fs
wave = sin.(2π*f_signal*t)

# window = hanning(nsamples)
window = blackman(nsamples)

wave_no_pad = wave .* window
h_no_pad = rfft(wave_no_pad)/(0.25nsamples)
freqs_no_pad = rfftfreq(length(wave_no_pad), fs)

npads = 2^12
wave_padded = [ zeros(npads); wave_no_pad ]
h_padded = rfft(wave_padded)/(0.25nsamples)
freqs_padded = rfftfreq(length(wave_padded), fs)

plot([], [], xlims=(20,20000), ylims=(-200,10), xscale=:log10, label="")
plot!(freqs_no_pad[2:end], dB.(h_no_pad[2:end]), label="With No Padding")
plot!(freqs_padded[2:end], dB.(h_padded[2:end]), label="Padded with Zeros")

 

[peng]

I am pretty sure the preculiar looking spectral (FFT magnitude) plots as mentioned in my post #4 were caused by mis-configured FFT analysis settings.

The scalloping in the THD spectrum plot looks like it was due to the analysis signal having zeros padded to it. Below are FFT magnitude plots generated with and without zero padding. Using more samples (i.e. increasing the FFT length) should also help with increasing the frequency resolution and thereby visually reduce the amount of spectral spreading (spectral leakage) in the plots.
View attachment 536797

Spoiler: Julia code used to create the above plots

# Reference ASR post: https://www.audiosciencereview.com/forum/index.php?threads/denon-avr-x3900h-preamp-measurements.71682/
using Plots
using FFTW
using DSP

dB(x) = 20.0log10(abs(x) + 5.0eps())

fs = 384000.0
nsamples = 2^16
f_signal = 1000.0

t = collect(0:nsamples-1)/fs
wave = sin.(2π*f_signal*t)

# window = hanning(nsamples)
window = blackman(nsamples)

wave_no_pad = wave .* window
h_no_pad = rfft(wave_no_pad)/(0.25nsamples)
freqs_no_pad = rfftfreq(length(wave_no_pad), fs)

npads = 2^12
wave_padded = [ zeros(npads); wave_no_pad ]
h_padded = rfft(wave_padded)/(0.25nsamples)
freqs_padded = rfftfreq(length(wave_padded), fs)

plot([], [], xlims=(20,20000), ylims=(-200,10), xscale=:log10, label="")
plot!(freqs_no_pad[2:end], dB.(h_no_pad[2:end]), label="With No Padding")
plot!(freqs_padded[2:end], dB.(h_padded[2:end]), label="Padded with Zeros")

Thanks, will try to learn and understand more, but in the mean time, could that explain a little, why such a large difference between his THD and THD+N, yet his Noise measurements wasn't bad at all, in fact he called it excellent?

 

[NTK]

Thanks, will try to learn and understand more, but in the mean time, could that explain a little, why such a large difference between his THD and THD+N, yet his Noise measurements wasn't bad at all, in fact he called it excellent?

I highly suspect the analyses were done with misconfigured settings. I don't think I can predict how they had affected the results, but I would not trust the numbers from the RMAA analyses in post #1. Some of them might be OK and some might not. OP said he was new to this so it is not too unusual that something had gone wrong.

 

[VintageFlanker]

Thanks for your efforts, @alex.schroedsen

Yet, I'm sorry to say that, but looking at these, most part if all measurements there seem flawed. By the shape of most waves, it looks like either sample frequency, I/O levels or time aren't synchronized/aligned properly (may probably be fixed in software domain). IMD, Noise and THD+N are clearly wrong, that's for sure. About the Crosstalk: it seems like your Cosmos is set in Mono-Mode.

 

[peng]

I highly suspect the analyses were done with misconfigured settings. I don't think I can predict how they had affected the results, but I would not trust the numbers from the RMAA analyses in post #1. Some of them might be OK and some might not. OP said he was new to this so it is not too unusual that something had gone wrong.

Not expecting you to predict how the results might have been affected, just wanted to know if I am the only one doubting at least some of the results presented at this point. Now I got that part answered, thanks again.

 

[peng]

When Amir review/measure the 3900, I hope he will include the one that he typically (almost always but not every time) would do, that is the SINAD vs frequency.
The AVR-X4800H and Cinema 40 did very bad relative to the AVR-X3700H, X3600H that use the AK4458 DAC IC, the rest of the parts such as the volume IC, opamps remain the same so the DAC is the one that really made a big difference, though as mentioned many time by other members, audibility may not be an issue.

I am using the Cinema 40 for reference, the numbers/graphs are almost the same as the AVR-X4800H but it costs a lot more and is clearly better build so imo it is a better reference:

For comparison, here's the AVR-X3700 that uses the AK4458 DAC chip:

We can see the significant deterioration going from the AVR-X3700H to the Cinema 40 in SINAD even in test frequencies at below 2,000 Hz, comparing the red curves.

Again, hopefully the AVR-X3900, Cinema 51/41 (the model numbers are assumed as nothing has been announced yet), will bring the SINAD vs Frequency back to where the Denon 3000/4000 AVRs performance level.

 

[EWL5]

I think everyone's waiting for Amir's review of the 3900 with bated breath!

I'll be curious about the Onkyo 2027 models as well but only to see if the "unintended 4 ohm mode" has been addressed. Can't imagine these hitting the streets until late 2027 or early 2028 if there are teething pains b/w Onkyo and ART.