Integrated Amplifier Comparison: $90 vs $9000

[greatestprateek]

Yes - but the differences should be definable according to objective principles and measurements. We don't have to enter the mug's game of continually swapping in different amplifiers and speakers until we accidentally hit on what we subjectively prefer that day.

With the greatest respect, this comment is terrible.

Music has evolved through the design of instruments and playing techniques, and the underlying principles of pitch and harmony, to be understood by humans. It matches our auditory skills, accordingly.

When it comes to a reproduction system though, there are two interrelated principles you don't seem to get. The first is that a reproduction system is not another musician. It should not contribute to the music so that we tell the difference that way: its job is to, ahem, reproduce. The music is the job of the musicians.

Secondly, and as a result, reproduction systems have to be judged in a different way to a musical instrument, and that part of the process is not something that humans are particularly well set up to do. So we can recognise patterns, pitch, voice inflection, and rhythm, but we don't do so well on tonality of instruments (it's a lot easier to tell two musicians apart, than the same professional musician playing two different, well made acoustic instruments), we don't get stereo imaging as well as we think: but we are incredibly confident we can tell these differences, and our brains when dealing with two sounds that are "supposed to be different" call upon sight, memory and anything else the brain has available to sort out what it is hearing.

The other thing to remember is that we don't need hi-fi to distinguish most or all of those "subtle features" in music you refer to. People do this with all those cheap earbuds and smart speakers and soundbars and all the other stuff we audiophiles tend to look down on. In fact, the hardest class of listening equipment to do this with is a poorly matched and set up full range stereo system. Too often we actually do damage.

Far from dismissing the auditory skills of humans, we recognise those skills in both their strengths and deficiencies. You would do well to understand this.

Just curious to know then, which headphone and speaker do you and all the people who liked this post, recommend as being as good as the best and endgame?

 

[Geert]

You cannot say that humans can create sophisticated music using their auditory skills but they cannot use those skills to judge audio equipment.

Yes you can, it's a different set of skills, each with different challenges and applicable laws of physics. It's not because there's some relation between music and sound that they are identical.

When they clearly use those skills to judge different instruments.

Differentiating between a violin and a trumpet does not require auditory skills beyond acknowledged audible thresholds.

I think you are vastly overvaluing few limited measurement techniques and using them to dismiss people’s auditory perception.

In fact measurement techniques outperform auditory capabilities. If you're not aware of that you have some studying to do.

 

[ahofer]

You cannot say that humans can create sophisticated music using their auditory skills but they cannot use those skills to judge audio equipment. When they clearly use those skills to judge different instruments. Its not about music affecting people, its about people evaluating music/sounds. I think you are vastly overvaluing few limited measurement techniques and using them to dismiss people’s auditory perception.

This is a non-argument in many different ways. There is sound, and there is perception. Equipment quality is about the sound, which exists in well-known dimensions and quality (frequency and amplitude), so that the artist's intentions will be rendered accurately.

Perception is indeed the stuff of humanity's crooked timber, and everyone here readily admits that the brain will react differently to different stimuli in difference environments. Composers meld the basic ingredients of sound into compositions with harmonic and melodic composition, production effects like saturation and mix, and, of course, their own interpretation of timing, pitch variance, and instrument feel to subtly shape those attributes. But all of those sounds are made up of measurable frequency and amplitude (phase being a by-product of those things). Different clarinets sound different due to their frequency and amplitude responding differently to wind and embouchure. The magic is the mixture and combination of measurable sounds and how the listener reacts to them, not some new unmeasurable sound ingredient for which you have no name and no evidence.

You are conflating very different things in an attempt to make mystery where there is little. The equipment does what is intended accurately, or it does not.

If you have evidence for some audible quality that is not measurable, or some way that human ears are more discriminating than measurement equipment, provide it. It would be groundbreaking. Otherwise stop muddying the waters with this nonsense.

 

[greatestprateek]

Differentiating between a violin and a trumpet does not require auditory skills beyond acknowledged audible thresholds.

Yes I didn’t phrase my point too well but you really think I am talking merely about differentiating different instruments. Obviously not. I am talking about mastering the nuances of different instruments, intuitively knowing how to make them each one of them at a time, sound subtly different = all the different sounds they can produce, holding onto some notes for longer, etc. One instrument can be made to produce sounds markedly quite different and having the ability to perceive what happened there. That’s what I mean. Using just one instrument, a musician can go along in a song. And even instruments of the same type can sound different. There is too much complexity here which has been instinctively mastered without the need to measure sound waves. People can tell the difference if a singer’s सुर Sur is off even slightly. You think they can’t tell if a kind of sound is being consistently and differently reproduced on separate audio equipment? Its naive to think they can’t.

 

[greatestprateek]

In fact measurement techniques outperform auditory capabilities. If you're not aware of that you have some studying to do.

The measurements techniques purported here are clearly in their infancy and not deep enough to find differences between sounds produced by audio equipment and perceived (differently) by people. There’s snake oil but there are also clear differences. If you show the same few graphs and say oh they measure flat/same, you have a lot of research to do before you can devise measurements that actually tell what’s going on? A lot of people are needed.

By the way, where is the International standard for measuring different audio equipment? Level matching voltage & test tones is not even scratching the surface of music(sound) complexity.

 

[ahofer]

There is too much complexity here which has been instinctively mastered without the need to measure sound waves. People can tell the difference if a singer’s सुर Sur is off even slightly. You think they can’t tell if a kind of sound is being consistently and differently reproduced on separate audio equipment? Its naive to think they can’t.

Again, the instruments are more accurate than ears, by a long shot. Our brains amplify and invent frequencies where we have hearing deficits, and you can convince people something sounds different just by showing them a different visual. It's naive to think that humans can consistently tell subtle differences in sound reproduction when their perception of sound is so easily warped by their brain.

 

[ahofer]

The measurements techniques purported here are clearly in their infancy and not deep enough to find differences between sounds produced by audio equipment and perceived (differently) by people. There’s snake oil but there are also clear differences. If you show the same few graphs and say oh they measure flat/same, you have a lot of research to do before you can devise measurements that actually tell what’s going on? A lot of people are needed.

By the way, where is the International standard for measuring different audio equipment? Level matching voltage & test tones is not even scratching the surface of music(sound) complexity.

I'm sorry, but everything you just said is wrong.

(@BDWoody I suspect this is heading towards the measurements thread)

 

[greatestprateek]

This is a non-argument in many different ways. There is sound, and there is perception. Equipment quality is about the sound, which exists in well-known dimensions and quality (frequency and amplitude), so that the artist's intentions will be rendered accurately.

Its a non argument if you miss the point. Is the music that people produce not sound too? Its a hugely sophisticated array of sounds. In fact I am afraid to inform you it is a lot more complex than fitting in your frequency and amplitude chart, because it also moves along the time domain. You cannot gloss over it. The basic perception of sounds requires no crooking or blurring by the brain, no poetic meanings attached to it. Sensory perception, is per se about just knowing that something happened, what happened. Humans can do that whats the big deal. What extra meaning we attach to it, or how we will go ahead and react to it, are even more different stories. Perception of sounds is first and foremost about whether that was audible or not. We noticed it or not. What extra things that means is a long story. You are going into that long story and saying its a non argument. I am saying music and its mere perception(notification/identification?) is very complex and requires the same kind of auditory skills that can be used to find differences between audio equipment.

 

[GXAlan]

The whole point of this post is that
1) I think I can hear a difference
2) I can measure the difference

So, the quality of our ears versus the quality of the measurements/instrumentation is not the question.

 

[audio_tony]

@GXAlan Do I understand correctly? You are measuring the amplifiers via a speaker?

So source -> amp -> speaker -> measurement mic. -> REW ?

If this is the case (and pardon me if this has already been mentioned, but I have (skim) read the thread.

Any differences you measure will be as a result of the interaction between the amplifier and speaker not so?

I know Amir's test include a graph showing frequency vs load impedance so this is to be expected? (in particular with some class D amps, but I believe that even some Class AB amps have some response changes with load)

 

[ahofer]

Is the music that people produce not sound too?

What did I say above when I talked about how music was made? Are you illiterate?

Its a hugely sophisticated array of sounds.

Yes, all comprised of the basic ingredients.

Perception of sounds is first and foremost about whether that was audible or not. We noticed it or not.

Yes, audibility is critical

You are going into that long story and saying its a non argument.

Indeed, because it was. Perception is separate from sound, as you seem to understand, yet your story is not what that implies.

because it also moves along the time domain

Frequency doesn't exist without time, it is implicit. Are you maintaining that audio equipment changes rhythms, or some other insane idea?

You are far too ignorant to argue with, you have no familiarity with measurements and the physics of sound, instruments and electronics, the limits of human hearing, the volatile subjectivity of human hearing, or even what constitutes musicianship. You just keep saying people respond to sound differently and somehow that makes the sound different and humans better judges of it. Both these ideas are false according to ALL current scientific evidence.

On to ignore with you, you are a fountain of ignorant babble.

 

[GXAlan]

@GXAlan Do I understand correctly? You are measuring the amplifiers via a speaker?

Yes. First two graphs are electrical measurements into a resistor.

The rest are acoustic measurements.

So source -> amp -> speaker -> measurement mic. -> REW ?

For the sweep yes. For the music, it was recorded using Audacity or Audition.

Any differences you measure will be as a result of the interaction between the amplifier and speaker not so?

This 1000%.

Take a look at my Rapahelite 300B SET review in my signature. It’s a long read because I was storytelling, but the speaker/amplifier interaction is WHY amplifiers that measure identically into a resistor may sound different in the real world.

My thesis is that not only is it different, but it meets the threshold of audibility.

I know Amir's test include a graph showing frequency vs load impedance so this is to be expected?

Yes and no. I agree it’s expected, but even in this thread, many people insist that if it was level matched, it would sound the same. My point is that if the amplifiers respond differently into loads, you cannot level match across the frequency spectrum. The other part is quantifying the difference.

I joined ASR right before the pandemic and went through an audiophile journey of trying so many different speakers and electronics over the last 4 years or so.

In the end:
- moved to $8k/pair active speakers for my audiophile setup (Meyer Sound Amie) and Dirac’d Bose 901 for my casual setup. I still have my Magnepan MG-III in storage.
- I still like SACDs and LP vinyl, with SACD being converted to PCM and LPs running through a tube phono stage
- have both the Yamaha CX-A5100 and Monolith HTP-1 as my primary line level products, selling my external DACs from Topping and my Marantz SA-10

 

[Galliardist]

Just curious to know then, which headphone and speaker do you and all the people who liked this post, recommend as being as good as the best and endgame?

I don’t see the relevance of this question unto what I wrote. And I suspect that what you are doing is trying to “prove ASR wrong” more than anything else.

Anyway, the question is, once again, wrong. There’s no best loudspeaker or headphone because there is no prefect one, and because the space for housing your proposed speakers is in the general case unique, as are most heads - inside as well as the ears, that will wear the headphones.

All we can do is rely on what we know to produce an appropriate compromise: and remember that sound is never going to be the only factor in any final decision.

 

[audio_tony]

My point is that if the amplifiers respond differently into loads, you cannot level match across the frequency spectrum.

Contrary to ASR opinion, I've always believed some amplifiers sound different when inserted into the same system.

Your testing seems to confirm this to be the case.

I recall reading some years ago (80's maybe) in a publication (forget which) where a similar conclusion was drawn, and the theory was that it was due to speaker / cable interaction with the amplifier feedback loop. From memory this conclusion was reached after testing various amplifiers with differing levels of feedback.

Sadly I don't remember the source of these tests.

 

[ahofer]

Contrary to ASR opinion, I've always believed some amplifiers sound different when inserted into the same system.

I think there are plenty of us who believe that, but specific to the phenomenon you discuss - impedance, output transformers, distortion profiles, etc. Measurable things, if a challenge to measure comprehensively.

On the other hand, many of us tend to view high output impedance, distortion, and load variability as design defects.

 

[audio_tony]

On the other hand, many of us tend to view high output impedance, distortion, and load variability as design defects.

I completely agree. However it would seem that even a competently designed solid state amplifier can suffer from load variability.

My feeling is that it's an inherent flaw in the design.

 

[ahofer]

competently designed solid state amplifier can suffer from load variability.

That would depend on the limitations to which it is designed. In any even, I think quite a few of the current class D designs available for a grand or less are more than sufficiently load-invariable. Managing the design to every bit of audiophoolery (see @pma ‘s thread on hypothetical electrostatic loads) cannot be the standard for competent.

 

[antcollinet]

I completely agree. However it would seem that even a competently designed solid state amplifier can suffer from load variability.

My feeling is that it's an inherent flaw in the design.

Perhaps - but generally only with extreme speakers run close to power limits.

Decent amps with low output impedance wont show (audible) load variablity with most speakers.

 

[totti1965]

I continue to hear differences between amplifiers that might be expected to be inaudibly different. That is, for two high-SINAD systems that I'm not running into clipping, I hear differences. I also believe in measurements. My perspective has been consistent: "Everything that can be measured may not be audible. Everything that is audible is measurable -- but you need to select the right measurement."

I think I'm a pretty open-minded audiophile. I've sold off higher-end, more prestigious speakers after discovering that a Bose 901 Series VI worked fortuitously well in my room. I've returned, with restocking fees and shipping expenses, new gear that I should have been biased to like from the novelty alone. Sometimes, I get to recover my losses by selling off an expensive product that is bested or matched by a lower priced product.

How does the $90 Fosi Audio V3 compare to the $9,000 Marantz PM-10 integrated amplifier?

Scenario:
I was lent a Fosi Audio V3 integrated amp with a 48V power supply with the request to run some measurements on it to make it was performing to spec. There wasn't any issue with it -- the owner knew that I have amplifier test gear and was curious. My 5W SINAD is 86.9 dB which is good enough to match @amirm 's measurements. I did my tests with an E1DA Cosmos ADC (Grade A) and Vishay-Dale 1% non-inductive resistors. I used my Marantz SA-10 as a signal generator, so the test instrumentation is nowhere as good as the APx555. I also didn't spend a lot of time with grounding. The measurements show that it's a good sample.

The Fosi Audio V3 on test
View attachment 302346

You can actually keep pushing this harder and harder and the SINAD keeps getting better.

The Marantz PM-10 has a SINAD that breaks the 100 dB barrier. This was done with a 20-22.05 kHz range instead of 20-20kHz, so it's probably in the 101 dB range.

Setting up my Subjective Tests
My reference integrated amplifier is the Marantz PM-10 which is built around a bridged HypeX NC500OEMs/SMPS600 design (so 4 NC500's + 2 SMPS600) along with Marantz's own pre-amp and buffer. My go-to passive bookshelf is the JBL XPL90 which has the benefit of being tested on the NFS here.

First step was volume matching. This was done with a MinDSP UMIK-2 placed at the main listening position.

Calibrating Volume
I used REW to calibrate levels. Per my conversations with @JohnPM , REW will send out pink noise from 500 Hz to 2000 Hz (a "speaker calibration") signal based upon my standardized sweep. The Marantz SA-10 outputs a fixed voltage and the Marantz PM-10 has 0.5 dB volume precision while the Fosi V3 has an analog logarithmic potentiometer.

With a -10 dBFS test signal in REW, I adjusted the volume reported by "Check Levels" to show 80 dB at the listening position with the speaker calibration pink noise. I then ran two sweeps above and below that level: +0 dBFS and -20 dBFS.

View attachment 302352

Volume matching is better than the 0.5 dB you can achieve with the Marantz's volume control. I have pretty good dexterity, and I am at the limit of what can be done by adjusting the Fosi V3 knob. You can see that the Fosi in green/red is slightly louder in the treble relative to the Marantz. This is just 1 dB of difference or less.

Once I had this level of volume matching, I did my listening tests by unplugging the Fosi V3 so that I wouldn't have to touch the volume dial.

Subjective Impressions
I started with the Fosi V3 and listened to Hotel California. My first impression was "wow. I didn't remember my XPL 90 having this kind of bass." When the vocals came in, I was disappointed and it sounded recessed or less natural. My initial thought was that I was just noticing the limitation of the speaker. I swapped to the Marantz PM-10. As soon as the track started, I felt as if the guitars were more natural. But the bass didn't impress. It wasn't anemic - it just didn't surprise and delight. When the voices came in, it sounded wonderful -- the sort of giddiness or smile that you probably got when you heard this track on your first premium system. The soundstage and vocals were noticeably better on the PM-10, but no one could reasonably say that it was 100X better. For me, it was the difference between "a wow and a smile" and "I can't complain".

Measurements
After my subjective impressions, I decided to try something new. I moved my UMIK-2 back to the listening position and recorded the first 3 minutes of Hotel California on each system. A 48 kHz sampling rate was used. I then brought the recordings into @pkane 's DeltaWave for analysis.

Spoiler: Deltawave Log

DeltaWave v2.0.8, 2023-07-30T21:12:49.7348193-07:00
Reference: PM-10_HC.flac[?] 8704000 samples 48000Hz 24bits, stereo, MD5=00
Comparison: FOSI_V3.flac[?] 8729280 samples 48000Hz 24bits, stereo, MD5=00
Settings:
Gain:True, Remove DC:True
Non-linear Gain EQ:False Non-linear Phase EQ: False
EQ FFT Size:65536, EQ Frequency Cut: 0Hz - 0Hz, EQ Threshold: -500dB
Correct Non-linearity: False
Correct Drift:True, Precision:30, Subsample Align:True
Non-Linear drift Correction:False
Upsample:False, Window:Kaiser
Spectrum Window:Kaiser, Spectrum Size:32768
Spectrogram Window:Hann, Spectrogram Size:4096, Spectrogram Steps:2048
Filter Type:FIR, window:Kaiser, taps:262144, minimum phase=False
Dither:False bits=0
Trim Silence:False
Enable Simple Waveform Measurement: False

Discarding Reference: Start=0.5s, End=0s
Discarding Comparison: Start=0.5s, End=0s

Initial peak values Reference: -10.514dB Comparison: -10.419dB
Initial RMS values Reference: -29.814dB Comparison: -29.761dB

Null Depth=11.544dB
X-Correlation offset: -8818 samples
Drift computation quality, #1: Good (5.78μs)


Trimmed 0 samples ( 0.00ms) front, 0 samples ( 0.00ms end)


Final peak values Reference: -10.514dB Comparison: -10.487dB
Final RMS values Reference: -29.814dB Comparison: -29.821dB

Gain= 0.0656dB (1.0076x) DC=0.00001 Phase offset=-183.706567ms (-8817.915 samples)
Difference (rms) = -58.08dB [-60.46dBA]
Correlated Null Depth=49.58dB [43.33dBA]
Clock drift: -0.27 ppm


Files are NOT a bit-perfect match (match=0.63%) at 16 bits
Files are NOT a bit-perfect match (match=0%) at 24 bits
Files match @ 49.9994% when reduced to 9.35 bits


---- Phase difference (full bandwidth): 12.3807379864222°
0-10kHz: 12.16°
0-20kHz: 12.35°
0-24kHz: 12.38°
Timing error (rms jitter): 15.3μs
PK Metric (step=400ms, overlap=50%):
RMS=-59.4dBr
Median=-59.6
Max=-53.9

99%: -55.25
75%: -57.85
50%: -59.6
25%: -62.34
1%: -73.91

gn=0.992477307560239, dc=9.05811550542576E-06, dr=-2.72459687718413E-07, of=-8817.9151922137

DONE!

Signature: 6ff930734cc68b8179e8896c8c5fb6df

RMS of the difference of spectra: -106.566381971716dB
DF Metric (step=400ms, overlap=0%):
Median=-28.8dB
Max=-6.6dB Min=-41dB

1% > -39.75dB
10% > -37.08dB
25% > -33.65dB
50% > -28.79dB
75% > -25.01dB
90% > -21.44dB
99% > -12.66dB

Linearity 22.8bits @ 0.5dB error

DeltaWave shows that my volume matching was as good as it gets. 0.05 dB difference in RMS. Deltawave corrects this to 0.007 dB.
Initial RMS values Reference: -29.814dB Comparison: -29.761dB

If we look at the aligned spectra, which is the comparison *after* DeltaWave has done its level/phase adjustments, the white (FOSI V3) does have more content below 30 Hz than the blue (Marantz PM-10).

View attachment 302349

The JBL XPL90 is a bookshelf speaker, so it cannot reproduce anything this low with meaningful SPLs. It's -30 dB at 20 Hz. That said, -30 dB still means ~55 dB when listening with 2.83V of power, and 50 dB is normal conversation. So maybe it's not so crazy to look at bass output this low. In the case of Hotel California, the drums enter at 30 seconds into the song and they are played largely in isolation -- so it would not be masked by other sounds in the music.

The difference between the two amps is a small 2 dB, but that's within the threshold of what is accepted as an audible difference.

What's interesting is that at high SPLs, the REW sweep showed no indication of this in frequency response. At lower SPL, the ripple around 20 Hz was something that I subconciously blew off as an artifact on the REW sweep. I didn't see it until writing up this post. The real-world bass content is more likely to be lower than 0 dBFS.

I believe this harmonizes the objectivist/subjectivist perspective.
If you only saw the 0 dBFS sweep, you would see that the two frequency responses are identical in the bass region. A subjectivist would distrust measurements because he/she would hear something different in the bass and not see a difference. An objectivist would dismiss the subjectivist claim of differences in bass as mere sighted bias or poorly controlled studies. However, by running the sweep at a different level, there may in fact be a difference. The objectivist is justified in saying that only frequency response changes can be audible and the subjectivist is justified in seeing that there is a measurement that correlates with what was heard, but required an atypical measurement.

What I also have learned, and hopefully not incorrectly, is to pay more attention to the lower frequencies and slope of the bass roll-off rather than just look for the -3 or -6 dB threshold. For years, there is always a claim of sealed vs. ported sound characteristics yet the measurements of "speed" are hard to explain since speed is dictated by the frequency of the audio. Purely speculating, maybe the difference in bass roll off slope is just as important as the single -6 dB point. I'd be curious to see a LS3/5a on the Klippel NFS to see how the bass measures. Part of the reason I ignored the oscillation on the REW sweep with the Fosi V3 is that I thought it was inconsequential on a bookshelf that didn't extend that deep.

So what about the difference in voices?
If we scrutinize the vocal region, there are differences in the recorded, un-corrected comparisons. Again, RMS volume mismatch is 0.05 dB different, but the biggest delta in this region is as high as 2 dB which could be audible.

View attachment 302364

Looking at the difference in spectra, you can see that the spectra is pretty consistent from 30 to 300 Hz, again suggesting that my measurements are reasonably done, but you do see a bigger difference as you move up the frequency range which is even bigger than what is seen in the bass region.
View attachment 302365

The XPL90 I used has this impedance/phase. I suppose the blip near 100 Hz in the delta of the spectra is an area that is a peak in both the phase and impedance, but I'm not seeing a pattern here and I don't know to how to calculate EPDR off a chart like this. Maybe this is harder to drive?

DF Metric and PK Metric are weighted metrics for performing the null test.
View attachment 302366

View attachment 302370

The PK Metric is looking at things throughout the entire spectrum, weighted perceptually, for a given point in time. The right image shows what results contributed the most to the calculation of the PK Metric and the two spikes are certainly within the region of the harmonics of the human voice.

The PK Metric of - 59.4 dBr isn't that high, but it may be due to PK Metric looking at things too broadly and being too strict. After all, we can accept that a drum solo with 2 dB of difference should be audible but the PK Metric doesn't really recognize this. I believe it's a moving average and if the drum solo is averaged with the guitar/voice, the PK Metric may assume that the bass is masked due to it being so much lower in dB than the "average" content in that sample bucket. We probably need some sort of ANOVA like calculation where any difference can result in detectable differences, and additional post-tests are needed to answer why.

Conclusion
There are two points to this post.

1. Two amplifiers with "good electrical performance" sounded different to my ears. While we know the Fosi V3 is load dependent, it translated into more of an issue that might be predicted. I was able to measure differences using a UMIK-2 that seem to correlate with my subjective experience. A "standard" sweep of 0 dBFS through a calibrated microphone would only have shown the high frequency differences of just 1 dB, which was not the biggest difference in my subjective impression. We talk about the advances in ADCs like the E1DA Cosmos ADC for measuring electrical performance, and I do believe that the UMIK-2 is a great tool for measuring speakers (at a fixed location).

2. I've compared a $90 and $9,000 MSRP product and presumably, you've read to this point. It says a lot about both products and the state of hi-fi. First, the $90 Fosi V3 shows the power of innovation and is a real showcase for Texas Instrument's Class D chip. It's sort of an incredible feat of engineering representing what is essentially a globally developed product. TI wouldn't be able to release a $90 integrated amplifier without a company like Fosi buying their chips and building the rest of the consumer product, nor would Fosi have been able to do this without the innovation of TI. When people work together, better things happen. My only comment on the Fosi is that the 32V power supply has UL listing while the 48V doesn't. While audio isn't the same as deep sea submersibles, I generally prefer to have certified products when the option is available or a product with a long pre-existing track record.

The $9,000 product is a clear demonstration of diminishing performance returns as you spend more money. On the other hand, much in the way one might spend a luxury vacation, there is an experience component to the audiophile hobby. The PM-10 would have delivered reliable Class D performance 7 years ago. Even if something like the Marantz Model 30 or even the 40n gives you the same audible experience within the power envelope, the luxury product delivered it earlier. Think ABS, driver assistance, and GPS -- luxury cars got it first.

As for the audible benefits of the luxury product? I certainly hear and prefer the difference, recognizing that I would never suggest that it is a "100X improvement". It was immediately noticed by me, but even 2 dB is subtle by any standard. I will leave it up to you to decide if the data presented is convincing enough that a difference between sound could in fact exist and that the difference could in fact reach audible thresholds. Neither amplifier hits the 120 dB SINAD threshold for absolute electrical transparency and the 5W electrical SINAD is ~87 vs ~101 dB.

The 2 dB Difference in the region of the human voice….. very interesting!
It would be nice to find loudspeakers with exactly 6 Ohms in the region between 8.000 and 20.000 Hz!
Maybe any differences would dissapear immediately!
6 Ohms is the value where the Fosi V3 would show his most linear performance.
If the high frequencies fit well - will the differences in the midrange will stay?
Most 2- or 3- way speakers have about 10 Ohms in the 8 kHz to 20 kHz region. And most Speakers are lacking 2 or 3 dB of energy in these extreme High frequencies. So maybe the Fosi with his + 1.0 dB at 20 kHz at 8 Ohm compensates some mistakes that the speakers do - and sounds for many people more „open“ or „natural“ than even 100 times more expensive amps!
By the way: in the SINAD Ranking list, the Fosi beats two amplifiers that even costs 1.000 times more……

 

[HarmonicTHD]

I continue to hear differences between amplifiers that might be expected to be inaudibly different. That is, for two high-SINAD systems that I'm not running into clipping, I hear differences. I also believe in measurements. My perspective has been consistent: "Everything that can be measured may not be audible. Everything that is audible is measurable -- but you need to select the right measurement."

I think I'm a pretty open-minded audiophile. I've sold off higher-end, more prestigious speakers after discovering that a Bose 901 Series VI worked fortuitously well in my room. I've returned, with restocking fees and shipping expenses, new gear that I should have been biased to like from the novelty alone. Sometimes, I get to recover my losses by selling off an expensive product that is bested or matched by a lower priced product.

How does the $90 Fosi Audio V3 compare to the $9,000 Marantz PM-10 integrated amplifier?

Scenario:
I was lent a Fosi Audio V3 integrated amp with a 48V power supply with the request to run some measurements on it to make it was performing to spec. There wasn't any issue with it -- the owner knew that I have amplifier test gear and was curious. My 5W SINAD is 86.9 dB which is good enough to match @amirm 's measurements. I did my tests with an E1DA Cosmos ADC (Grade A) and Vishay-Dale 1% non-inductive resistors. I used my Marantz SA-10 as a signal generator, so the test instrumentation is nowhere as good as the APx555. I also didn't spend a lot of time with grounding. The measurements show that it's a good sample.

The Fosi Audio V3 on test
View attachment 302346

You can actually keep pushing this harder and harder and the SINAD keeps getting better.

The Marantz PM-10 has a SINAD that breaks the 100 dB barrier. This was done with a 20-22.05 kHz range instead of 20-20kHz, so it's probably in the 101 dB range.

Setting up my Subjective Tests
My reference integrated amplifier is the Marantz PM-10 which is built around a bridged HypeX NC500OEMs/SMPS600 design (so 4 NC500's + 2 SMPS600) along with Marantz's own pre-amp and buffer. My go-to passive bookshelf is the JBL XPL90 which has the benefit of being tested on the NFS here.

First step was volume matching. This was done with a MinDSP UMIK-2 placed at the main listening position.

Calibrating Volume
I used REW to calibrate levels. Per my conversations with @JohnPM , REW will send out pink noise from 500 Hz to 2000 Hz (a "speaker calibration") signal based upon my standardized sweep. The Marantz SA-10 outputs a fixed voltage and the Marantz PM-10 has 0.5 dB volume precision while the Fosi V3 has an analog logarithmic potentiometer.

With a -10 dBFS test signal in REW, I adjusted the volume reported by "Check Levels" to show 80 dB at the listening position with the speaker calibration pink noise. I then ran two sweeps above and below that level: +0 dBFS and -20 dBFS.

View attachment 302352

Volume matching is better than the 0.5 dB you can achieve with the Marantz's volume control. I have pretty good dexterity, and I am at the limit of what can be done by adjusting the Fosi V3 knob. You can see that the Fosi in green/red is slightly louder in the treble relative to the Marantz. This is just 1 dB of difference or less.

Once I had this level of volume matching, I did my listening tests by unplugging the Fosi V3 so that I wouldn't have to touch the volume dial.

Subjective Impressions
I started with the Fosi V3 and listened to Hotel California. My first impression was "wow. I didn't remember my XPL 90 having this kind of bass." When the vocals came in, I was disappointed and it sounded recessed or less natural. My initial thought was that I was just noticing the limitation of the speaker. I swapped to the Marantz PM-10. As soon as the track started, I felt as if the guitars were more natural. But the bass didn't impress. It wasn't anemic - it just didn't surprise and delight. When the voices came in, it sounded wonderful -- the sort of giddiness or smile that you probably got when you heard this track on your first premium system. The soundstage and vocals were noticeably better on the PM-10, but no one could reasonably say that it was 100X better. For me, it was the difference between "a wow and a smile" and "I can't complain".

Measurements
After my subjective impressions, I decided to try something new. I moved my UMIK-2 back to the listening position and recorded the first 3 minutes of Hotel California on each system. A 48 kHz sampling rate was used. I then brought the recordings into @pkane 's DeltaWave for analysis.

Spoiler: Deltawave Log

DeltaWave v2.0.8, 2023-07-30T21:12:49.7348193-07:00
Reference: PM-10_HC.flac[?] 8704000 samples 48000Hz 24bits, stereo, MD5=00
Comparison: FOSI_V3.flac[?] 8729280 samples 48000Hz 24bits, stereo, MD5=00
Settings:
Gain:True, Remove DC:True
Non-linear Gain EQ:False Non-linear Phase EQ: False
EQ FFT Size:65536, EQ Frequency Cut: 0Hz - 0Hz, EQ Threshold: -500dB
Correct Non-linearity: False
Correct Drift:True, Precision:30, Subsample Align:True
Non-Linear drift Correction:False
Upsample:False, Window:Kaiser
Spectrum Window:Kaiser, Spectrum Size:32768
Spectrogram Window:Hann, Spectrogram Size:4096, Spectrogram Steps:2048
Filter Type:FIR, window:Kaiser, taps:262144, minimum phase=False
Dither:False bits=0
Trim Silence:False
Enable Simple Waveform Measurement: False

Discarding Reference: Start=0.5s, End=0s
Discarding Comparison: Start=0.5s, End=0s

Initial peak values Reference: -10.514dB Comparison: -10.419dB
Initial RMS values Reference: -29.814dB Comparison: -29.761dB

Null Depth=11.544dB
X-Correlation offset: -8818 samples
Drift computation quality, #1: Good (5.78μs)


Trimmed 0 samples ( 0.00ms) front, 0 samples ( 0.00ms end)


Final peak values Reference: -10.514dB Comparison: -10.487dB
Final RMS values Reference: -29.814dB Comparison: -29.821dB

Gain= 0.0656dB (1.0076x) DC=0.00001 Phase offset=-183.706567ms (-8817.915 samples)
Difference (rms) = -58.08dB [-60.46dBA]
Correlated Null Depth=49.58dB [43.33dBA]
Clock drift: -0.27 ppm


Files are NOT a bit-perfect match (match=0.63%) at 16 bits
Files are NOT a bit-perfect match (match=0%) at 24 bits
Files match @ 49.9994% when reduced to 9.35 bits


---- Phase difference (full bandwidth): 12.3807379864222°
0-10kHz: 12.16°
0-20kHz: 12.35°
0-24kHz: 12.38°
Timing error (rms jitter): 15.3μs
PK Metric (step=400ms, overlap=50%):
RMS=-59.4dBr
Median=-59.6
Max=-53.9

99%: -55.25
75%: -57.85
50%: -59.6
25%: -62.34
1%: -73.91

gn=0.992477307560239, dc=9.05811550542576E-06, dr=-2.72459687718413E-07, of=-8817.9151922137

DONE!

Signature: 6ff930734cc68b8179e8896c8c5fb6df

RMS of the difference of spectra: -106.566381971716dB
DF Metric (step=400ms, overlap=0%):
Median=-28.8dB
Max=-6.6dB Min=-41dB

1% > -39.75dB
10% > -37.08dB
25% > -33.65dB
50% > -28.79dB
75% > -25.01dB
90% > -21.44dB
99% > -12.66dB

Linearity 22.8bits @ 0.5dB error

DeltaWave shows that my volume matching was as good as it gets. 0.05 dB difference in RMS. Deltawave corrects this to 0.007 dB.
Initial RMS values Reference: -29.814dB Comparison: -29.761dB

If we look at the aligned spectra, which is the comparison *after* DeltaWave has done its level/phase adjustments, the white (FOSI V3) does have more content below 30 Hz than the blue (Marantz PM-10).

View attachment 302349

The JBL XPL90 is a bookshelf speaker, so it cannot reproduce anything this low with meaningful SPLs. It's -30 dB at 20 Hz. That said, -30 dB still means ~55 dB when listening with 2.83V of power, and 50 dB is normal conversation. So maybe it's not so crazy to look at bass output this low. In the case of Hotel California, the drums enter at 30 seconds into the song and they are played largely in isolation -- so it would not be masked by other sounds in the music.

The difference between the two amps is a small 2 dB, but that's within the threshold of what is accepted as an audible difference.

What's interesting is that at high SPLs, the REW sweep showed no indication of this in frequency response. At lower SPL, the ripple around 20 Hz was something that I subconciously blew off as an artifact on the REW sweep. I didn't see it until writing up this post. The real-world bass content is more likely to be lower than 0 dBFS.

I believe this harmonizes the objectivist/subjectivist perspective.
If you only saw the 0 dBFS sweep, you would see that the two frequency responses are identical in the bass region. A subjectivist would distrust measurements because he/she would hear something different in the bass and not see a difference. An objectivist would dismiss the subjectivist claim of differences in bass as mere sighted bias or poorly controlled studies. However, by running the sweep at a different level, there may in fact be a difference. The objectivist is justified in saying that only frequency response changes can be audible and the subjectivist is justified in seeing that there is a measurement that correlates with what was heard, but required an atypical measurement.

What I also have learned, and hopefully not incorrectly, is to pay more attention to the lower frequencies and slope of the bass roll-off rather than just look for the -3 or -6 dB threshold. For years, there is always a claim of sealed vs. ported sound characteristics yet the measurements of "speed" are hard to explain since speed is dictated by the frequency of the audio. Purely speculating, maybe the difference in bass roll off slope is just as important as the single -6 dB point. I'd be curious to see a LS3/5a on the Klippel NFS to see how the bass measures. Part of the reason I ignored the oscillation on the REW sweep with the Fosi V3 is that I thought it was inconsequential on a bookshelf that didn't extend that deep.

So what about the difference in voices?
If we scrutinize the vocal region, there are differences in the recorded, un-corrected comparisons. Again, RMS volume mismatch is 0.05 dB different, but the biggest delta in this region is as high as 2 dB which could be audible.

View attachment 302364

Looking at the difference in spectra, you can see that the spectra is pretty consistent from 30 to 300 Hz, again suggesting that my measurements are reasonably done, but you do see a bigger difference as you move up the frequency range which is even bigger than what is seen in the bass region.
View attachment 302365

The XPL90 I used has this impedance/phase. I suppose the blip near 100 Hz in the delta of the spectra is an area that is a peak in both the phase and impedance, but I'm not seeing a pattern here and I don't know to how to calculate EPDR off a chart like this. Maybe this is harder to drive?

DF Metric and PK Metric are weighted metrics for performing the null test.
View attachment 302366

View attachment 302370

The PK Metric is looking at things throughout the entire spectrum, weighted perceptually, for a given point in time. The right image shows what results contributed the most to the calculation of the PK Metric and the two spikes are certainly within the region of the harmonics of the human voice.

The PK Metric of - 59.4 dBr isn't that high, but it may be due to PK Metric looking at things too broadly and being too strict. After all, we can accept that a drum solo with 2 dB of difference should be audible but the PK Metric doesn't really recognize this. I believe it's a moving average and if the drum solo is averaged with the guitar/voice, the PK Metric may assume that the bass is masked due to it being so much lower in dB than the "average" content in that sample bucket. We probably need some sort of ANOVA like calculation where any difference can result in detectable differences, and additional post-tests are needed to answer why.

Conclusion
There are two points to this post.

1. Two amplifiers with "good electrical performance" sounded different to my ears. While we know the Fosi V3 is load dependent, it translated into more of an issue that might be predicted. I was able to measure differences using a UMIK-2 that seem to correlate with my subjective experience. A "standard" sweep of 0 dBFS through a calibrated microphone would only have shown the high frequency differences of just 1 dB, which was not the biggest difference in my subjective impression. We talk about the advances in ADCs like the E1DA Cosmos ADC for measuring electrical performance, and I do believe that the UMIK-2 is a great tool for measuring speakers (at a fixed location).

2. I've compared a $90 and $9,000 MSRP product and presumably, you've read to this point. It says a lot about both products and the state of hi-fi. First, the $90 Fosi V3 shows the power of innovation and is a real showcase for Texas Instrument's Class D chip. It's sort of an incredible feat of engineering representing what is essentially a globally developed product. TI wouldn't be able to release a $90 integrated amplifier without a company like Fosi buying their chips and building the rest of the consumer product, nor would Fosi have been able to do this without the innovation of TI. When people work together, better things happen. My only comment on the Fosi is that the 32V power supply has UL listing while the 48V doesn't. While audio isn't the same as deep sea submersibles, I generally prefer to have certified products when the option is available or a product with a long pre-existing track record.

The $9,000 product is a clear demonstration of diminishing performance returns as you spend more money. On the other hand, much in the way one might spend a luxury vacation, there is an experience component to the audiophile hobby. The PM-10 would have delivered reliable Class D performance 7 years ago. Even if something like the Marantz Model 30 or even the 40n gives you the same audible experience within the power envelope, the luxury product delivered it earlier. Think ABS, driver assistance, and GPS -- luxury cars got it first.

As for the audible benefits of the luxury product? I certainly hear and prefer the difference, recognizing that I would never suggest that it is a "100X improvement". It was immediately noticed by me, but even 2 dB is subtle by any standard. I will leave it up to you to decide if the data presented is convincing enough that a difference between sound could in fact exist and that the difference could in fact reach audible thresholds. Neither amplifier hits the 120 dB SINAD threshold for absolute electrical transparency and the 5W electrical SINAD is ~87 vs ~101 dB.

Great effort. Thank you. Sorry I am just seeing this.

Question: Did you ever get around estimating your measurement error? (Sorry if I missed it, as I skipped through some of the pages).

If not, maybe next time run 10 sweeps with one amp, flip run 10 sweeps, flip back and another 10. Do it ideally three times. See if the curves still fall on top of each other for each amp. Or if they „blur“ and to what extent. Then compare with this

Would be very interesting and would prove or disprove if the difference you measured and later heard here are due to measurement/ recording variations or if they are indeed due to the amps.

(There is even a method behind it, but you don’t need to do it that academically to get an idea. It’s called MSA and the results (GRandR) show you which measurement errors are caused by the measurement device and which by the operator. It is used worldwide and is an quasi industry standard).