j-curve
100+ Head-Fier
- Joined
- Apr 15, 2002
- Posts
- 489
- Likes
- 11
A couple of weeks ago I was listening to the sample disc from Gramophone magazine (February) when I noticed something odd from my trusty portable CD player - a subtle effect but definitely something there, which if I had to "give it a name" would be noise pumping.
The track was the 2nd movement of the Schumann piano concerto performed by Grimaud/Salonen/Staatskapelle Dresden(2006). You should be able to hear a low-res sample by clicking on the Intermezzo Andantino Grazioso. It starts with four gently rising chords on the piano, answered by the orchestra, then repeated. Very nice... unless someone is grinding pepper with the music, or twisting a single maraca between open palms. [They should twist a slice of lime instead.
]
What the heck is that? A quick check of a couple of other CD players around the place and the finger was pointing back at the NE900.
Surely not! I've come to trust the little Sony and in the past it has shown up well against component and mini-system CD players. Of course, that was always with the HD580, whereas this was with the HD215.
Blame the 'phones? Well, it's not as if they were rattling. They do have more top-end than the HD580 but more to the point (as I would discover) they are much more sensitive. Although equally efficient at 97dB/mW, the low impedance of the HD215 makes them 10dB more sensitive than the HD580 for a given voltage. In practical terms that means they're about twice as loud, unless you turn the volume down.
Plugging through an amp to keep the final level in the headphones almost constant, I was able to listen for whether the NE900's sound quality changes depending on its volume setting. Indeed it does:-
30/30 "perfect"
25/30 first hint of noise pumping, only noticeable if you're listening closely for it
20/30 someone using sandpaper lightly at a distance - easily heard, but music still listenable
15/30 affecting the orchestral strings now too, not just the piano
10/30 dragging your thongs through gravel in time with the music.
Of course you don't need the Schumann piano concerto to hear it, but piano is good, as is acoustic guitar if not too distorted or compressed. Once the chaos of each transient has subsided, the noise gets pumped in over the decay, so instead of "with the music" I should perhaps say "lagging behind the beat".
Could this anomaly be due to a poorly implemented digital volume control (as opposed to a digitally controlled analogue volume control)? Time to run the ruler over the NE900 and see what's happening.
I generated some test tones with Audacity and burned them to a CD-RW, 30 seconds of 100Hz, 1kHz, 10kHz and white noise, each at -6dB FS (full scale) and then at 0dB FS, with 5 second fades 'cos I never want to get hit in the ear with a FS test tone. Here are my measurements of the D-NE900 (Japanese domestic version):-
Output Impedance (headphone)
Sony spec: 16 ohms
Measured: 6 ohms
Output Power (headphone)
Sony spec: 5mW per channel into 16 ohms
Measured: 5mW (cleanly) into 10 ohms at volume=24/30
3.7mW (cleanly) into 16 ohms at volume=24/30
3.5mW (cleanly) into 32 ohms at volume=25/30
With a 1kHz sine wave, I couldn't quite achieve the specification. Maybe JEITA allows a different test signal, an external battery, or doesn't insist that full power be delivered cleanly.
Output Level (line)
Sony spec: 0.7Vrms at 47kilohms
Measured: 710mV (@1kHz FS) at only 21 ohms(!)
This is equivalent to 2V peak-peak, i.e. standard CD level. Probably shouldn't get too excited about the low impedance as there's no current to back it up, only 100uA peak according to the specification, which is insufficient to drive headphones.
Output Level (headphone)
Sony spec: not stated
Measured: Full scale output at volume=30/30 with no load would be 912mV rms, however this is not achievable due to clipping. More about that below, but interesting to note that the internal headphone amplifier provides a mild 2dB of gain above the unit's line level.
Volume Steps
vol 30/30 0.0 dB [reference level = 912mV * Zload/(Zload + 6), where Zload=impedance of headphones]
vol 29/30 -1.5 dB
vol 28/30 -3.0 dB
vol 27/30 -4.5 dB
vol 26/30 -6.1 dB
vol 25/30 -7.6 dB
vol 24/30 -9.0 dB
vol 23/30 -10.6 dB
vol 22/30 -11.7 dB
vol 21/30 -12.7 dB
vol 20/30 -14.3 dB
vol 19/30 -15.8 dB
vol 18/30 -17.1 dB
vol 17/30 -18.4 dB
vol 16/30 -19.7 dB
vol 15/30 -21.1 dB
vol 14/30 -22.5 dB
vol 13/30 -23.9 dB
vol 12/30 -25.6 dB
vol 11/30 -27.2 dB
vol 10/30 -28.8 dB
vol 9/30 -30.6 dB
vol 8/30 -32.1 dB
vol 7/30 -33.9 dB
vol 6/30 -36.1 dB
vol 5/30 -37.9 dB
vol 4/30 -39.6 dB
vol 3/30 -41.7 dB
vol 2/30 -45.6 dB
vol 1/30 -52.9 dB
This chart might help when comparing the relative sensitivity of different headphones. I hope someone finds it useful anyway.
[This would have been much easier if the NE900 displayed the volume as a number in addition to the trendy bar graph!] For volumes above 10/30 you can fairly reliably assume 1.5dB per step.
The other implication of these numbers is that if the NE900 has a digital volume control, then each 6dB drop in level corresponds to a loss of one bit of resolution, i.e. with volume 26/30 you're listening to 15-bit audio, vol=21/30 gives 14-bit, etc. If you have sensitive headphones and a loud CD you could easily have the volume set at 13/30 and effectively be getting 12-bit audio. Is that acceptable?
Analogue Clipping
The first astonishing thing that struck me during this testing is that the NE900 can't reproduce a full scale 1kHz sine wave at 30/30 volume, even with no load! Clipping starts about 7dB below full output, so if the volume is 26/30 or higher you have distortion. Plugging in some 32 ohm headphones makes things worse, but not by much - clipping then starts at -8dB. This would seem to be a serious flaw, but if the CD's you play aren't compressed too tightly into the top 8dB of dynamic range then you might not notice.
THD
Measuring one 16-bit device with another 16-bit device is pure buffoonery, so let's "measure" the total harmonic distortion of the NE900 with a computer soundcard, just for laughs.
To convert to the more familiar voltage ratio:-
-20dB = 10% THD
-40dB = 1% THD
-60dB = 0.1% THD
-80dB = 0.01% THD
This is indeed a graph of THD, but the big question is, the THD of what, CD player or soundcard? [My soundcard is actually a Cirrus Logic CS4237B chip built onto the motherboard. The chip comes in two versions, one with a typical input THD of 0.05%(-66dB) and another with 0.006%(-84dB). Looks like I got the better one.
]
Obviously, the sharp rise in distortion at high volumes (right hand side of graph) is due to analogue clipping in the output stage of the NE900. The general floor just below -80dB is no doubt the limitation of the soundcard, but insofar as THD is cumulative, it also means that the CD player is no worse than 0.01% THD there, a remarkable achievement for a portable.
To confirm my suspicion that the rise in THD at lower volumes is the fault of the Sony and not the soundcard, I used a potentiometer (analog volume control) to generate clean low-level signals to feed the soundcard. All other conditions were identical, with the input volume controls (line and master) maxxed out. As the green trace shows, the soundcard is also near its limits but sufficiently below the blue line that we can be sure the NE900 is responsible for the bulk of the distortion in the blue trace.
Summary
- The NE900 utilises a digital volume control which reduces the effective bit depth as the volume is reduced, leading to impaired resolution and increased harmonic distortion. This will affect sound quality at volumes of 20/30 or below. Listeners of acoustic recordings on sensitive headphones (more than 110dB/V) are most likely to notice.
- The NE900 headphone output clips at about 8dB below full scale, which will affect listeners of heavily compressed music on headphones with low sensitivity, such that the volume control is 25/30 or above. [May not apply to European model].
- However, listeners of uncompressed music on low sensitivity (less than 105dB/V) headphones may escape both these problems, because clipping will be rare and the digital volume control artefacts will be inaudible at high volume settings. [This was me with the HD580].
- Otherwise, the general rule that portables sound better with external amplification will apply to the NE900. The good news is that the line out of the NE900 is superb (IMO).
Question
The THD graph has me puzzled - it's just too good (ignoring the clipping). I don't expect to be able to hear 0.01% THD at volume 15/30, and yet the anomaly I'm hearing is obvious. THD is probably the wrong measurement. It doesn't even sound like harmonic distortion, more like broadband noise, a kind of scratching. It's dynamic too, heard on gradual decays, not so much in the steady state (sustain). Can someone suggest a better measurement?
And a final question, just to be safe. Does anyone else hear this digital volume control pepper grinding or am I imagining things?
The track was the 2nd movement of the Schumann piano concerto performed by Grimaud/Salonen/Staatskapelle Dresden(2006). You should be able to hear a low-res sample by clicking on the Intermezzo Andantino Grazioso. It starts with four gently rising chords on the piano, answered by the orchestra, then repeated. Very nice... unless someone is grinding pepper with the music, or twisting a single maraca between open palms. [They should twist a slice of lime instead.
What the heck is that? A quick check of a couple of other CD players around the place and the finger was pointing back at the NE900.
Blame the 'phones? Well, it's not as if they were rattling. They do have more top-end than the HD580 but more to the point (as I would discover) they are much more sensitive. Although equally efficient at 97dB/mW, the low impedance of the HD215 makes them 10dB more sensitive than the HD580 for a given voltage. In practical terms that means they're about twice as loud, unless you turn the volume down.
Plugging through an amp to keep the final level in the headphones almost constant, I was able to listen for whether the NE900's sound quality changes depending on its volume setting. Indeed it does:-
30/30 "perfect"
25/30 first hint of noise pumping, only noticeable if you're listening closely for it
20/30 someone using sandpaper lightly at a distance - easily heard, but music still listenable
15/30 affecting the orchestral strings now too, not just the piano
10/30 dragging your thongs through gravel in time with the music.
Of course you don't need the Schumann piano concerto to hear it, but piano is good, as is acoustic guitar if not too distorted or compressed. Once the chaos of each transient has subsided, the noise gets pumped in over the decay, so instead of "with the music" I should perhaps say "lagging behind the beat".
Could this anomaly be due to a poorly implemented digital volume control (as opposed to a digitally controlled analogue volume control)? Time to run the ruler over the NE900 and see what's happening.
I generated some test tones with Audacity and burned them to a CD-RW, 30 seconds of 100Hz, 1kHz, 10kHz and white noise, each at -6dB FS (full scale) and then at 0dB FS, with 5 second fades 'cos I never want to get hit in the ear with a FS test tone. Here are my measurements of the D-NE900 (Japanese domestic version):-
Output Impedance (headphone)
Sony spec: 16 ohms
Measured: 6 ohms
Output Power (headphone)
Sony spec: 5mW per channel into 16 ohms
Measured: 5mW (cleanly) into 10 ohms at volume=24/30
3.7mW (cleanly) into 16 ohms at volume=24/30
3.5mW (cleanly) into 32 ohms at volume=25/30
With a 1kHz sine wave, I couldn't quite achieve the specification. Maybe JEITA allows a different test signal, an external battery, or doesn't insist that full power be delivered cleanly.
Output Level (line)
Sony spec: 0.7Vrms at 47kilohms
Measured: 710mV (@1kHz FS) at only 21 ohms(!)
This is equivalent to 2V peak-peak, i.e. standard CD level. Probably shouldn't get too excited about the low impedance as there's no current to back it up, only 100uA peak according to the specification, which is insufficient to drive headphones.
Output Level (headphone)
Sony spec: not stated
Measured: Full scale output at volume=30/30 with no load would be 912mV rms, however this is not achievable due to clipping. More about that below, but interesting to note that the internal headphone amplifier provides a mild 2dB of gain above the unit's line level.
Volume Steps
vol 30/30 0.0 dB [reference level = 912mV * Zload/(Zload + 6), where Zload=impedance of headphones]
vol 29/30 -1.5 dB
vol 28/30 -3.0 dB
vol 27/30 -4.5 dB
vol 26/30 -6.1 dB
vol 25/30 -7.6 dB
vol 24/30 -9.0 dB
vol 23/30 -10.6 dB
vol 22/30 -11.7 dB
vol 21/30 -12.7 dB
vol 20/30 -14.3 dB
vol 19/30 -15.8 dB
vol 18/30 -17.1 dB
vol 17/30 -18.4 dB
vol 16/30 -19.7 dB
vol 15/30 -21.1 dB
vol 14/30 -22.5 dB
vol 13/30 -23.9 dB
vol 12/30 -25.6 dB
vol 11/30 -27.2 dB
vol 10/30 -28.8 dB
vol 9/30 -30.6 dB
vol 8/30 -32.1 dB
vol 7/30 -33.9 dB
vol 6/30 -36.1 dB
vol 5/30 -37.9 dB
vol 4/30 -39.6 dB
vol 3/30 -41.7 dB
vol 2/30 -45.6 dB
vol 1/30 -52.9 dB
This chart might help when comparing the relative sensitivity of different headphones. I hope someone finds it useful anyway.
The other implication of these numbers is that if the NE900 has a digital volume control, then each 6dB drop in level corresponds to a loss of one bit of resolution, i.e. with volume 26/30 you're listening to 15-bit audio, vol=21/30 gives 14-bit, etc. If you have sensitive headphones and a loud CD you could easily have the volume set at 13/30 and effectively be getting 12-bit audio. Is that acceptable?
Analogue Clipping
The first astonishing thing that struck me during this testing is that the NE900 can't reproduce a full scale 1kHz sine wave at 30/30 volume, even with no load! Clipping starts about 7dB below full output, so if the volume is 26/30 or higher you have distortion. Plugging in some 32 ohm headphones makes things worse, but not by much - clipping then starts at -8dB. This would seem to be a serious flaw, but if the CD's you play aren't compressed too tightly into the top 8dB of dynamic range then you might not notice.
THD
Measuring one 16-bit device with another 16-bit device is pure buffoonery, so let's "measure" the total harmonic distortion of the NE900 with a computer soundcard, just for laughs.
To convert to the more familiar voltage ratio:-
-20dB = 10% THD
-40dB = 1% THD
-60dB = 0.1% THD
-80dB = 0.01% THD
This is indeed a graph of THD, but the big question is, the THD of what, CD player or soundcard? [My soundcard is actually a Cirrus Logic CS4237B chip built onto the motherboard. The chip comes in two versions, one with a typical input THD of 0.05%(-66dB) and another with 0.006%(-84dB). Looks like I got the better one.
Obviously, the sharp rise in distortion at high volumes (right hand side of graph) is due to analogue clipping in the output stage of the NE900. The general floor just below -80dB is no doubt the limitation of the soundcard, but insofar as THD is cumulative, it also means that the CD player is no worse than 0.01% THD there, a remarkable achievement for a portable.
To confirm my suspicion that the rise in THD at lower volumes is the fault of the Sony and not the soundcard, I used a potentiometer (analog volume control) to generate clean low-level signals to feed the soundcard. All other conditions were identical, with the input volume controls (line and master) maxxed out. As the green trace shows, the soundcard is also near its limits but sufficiently below the blue line that we can be sure the NE900 is responsible for the bulk of the distortion in the blue trace.
Summary
- The NE900 utilises a digital volume control which reduces the effective bit depth as the volume is reduced, leading to impaired resolution and increased harmonic distortion. This will affect sound quality at volumes of 20/30 or below. Listeners of acoustic recordings on sensitive headphones (more than 110dB/V) are most likely to notice.
- The NE900 headphone output clips at about 8dB below full scale, which will affect listeners of heavily compressed music on headphones with low sensitivity, such that the volume control is 25/30 or above. [May not apply to European model].
- However, listeners of uncompressed music on low sensitivity (less than 105dB/V) headphones may escape both these problems, because clipping will be rare and the digital volume control artefacts will be inaudible at high volume settings. [This was me with the HD580].
- Otherwise, the general rule that portables sound better with external amplification will apply to the NE900. The good news is that the line out of the NE900 is superb (IMO).
Question
The THD graph has me puzzled - it's just too good (ignoring the clipping). I don't expect to be able to hear 0.01% THD at volume 15/30, and yet the anomaly I'm hearing is obvious. THD is probably the wrong measurement. It doesn't even sound like harmonic distortion, more like broadband noise, a kind of scratching. It's dynamic too, heard on gradual decays, not so much in the steady state (sustain). Can someone suggest a better measurement?
And a final question, just to be safe. Does anyone else hear this digital volume control pepper grinding or am I imagining things?
