charleski
Major Contributor
This is a review and measurement of the venerable Musical Fidelity XCans (V1) headphone amplifier. This was first released in 1998 for the surprisingly-affordable price of £129 (I've attached a flyer and pricelist from 1998 below). The '90s were marked by the renaissance of pricey valve electronics in hifiland and this product offered audiophiles a chance to experience the magic of chubes without having to fork out $60k in 1990 dollars for an Audio Note Ongaku. I suspect I bought the unit on test in 2000. Since I don't have any panthers my dog lent a squeaky toy for the picture.
At the time the XCans was the subject of a bit of a modding fad, and you can still find a site selling capacitor kits here. As can be seen from the pictures, I tried my hand at this, switching the electrolytics for fancier 'audio' grade versions, bypassing them with polystyrene caps and changing the ceramic feedback compensation caps for polystyrene. Oddly enough, I couldn't hear the slightest difference after making this 'upgrade' ... When I exhumed the device from storage last year I noticed that one of the capacitors in the power supply was bulging, so replaced them all with 105C Panasonics. I added clip-on heatsinks to the BD139s (TR102) as there were signs of heat-stress on the PCB and drilled some holes in the casing to allow a small amount of airflow (as sold there was no ventilation at all).
The amp's transformer is in a wall-wart which feeds it 12VAC via a barrel connector. The valves are run as cathode-followers with an active transistor load and an opamp gain stage. It seems to be a unique hybrid design and I can't find anything similar in the resources on valve amplifier design. It's certainly very different to a White cathode-follower, which is the usual starting point for active-load CFs. One flaw that I found while tinkering with it is that if the valves aren't conducting it places 30VDC across the output. This also happens when it's powered up, as the PSU capacitors charge up before the heaters have had time to raise the cathodes to the correct temperature. So I added a timer relay to the input to the main diode bridge which turns it on 40 seconds after the heaters have received power. Fixing this properly would require a thorough redesign, but this works as a stopgap.
The output impedance is 49ohms, largely as a result of R115, the 47ohm current-limiting resistor on the output. When it was introduced it was championed as a good match for 300ohm Sennheiser HD600s, but I've found it works quite well with my ultra-low impedance (18ohm) planars, though this may be because they don't present a reactive load. Obviously with low-impedance headphones the majority of the power produced is actually being dissipated in the output resistor.
Measurements
I recently resurrected my 15-yr-old EMU 1616m audio interface running off an antique laptop with the requisite PCMCIA slot. While old, this boasted class-leading specs back in the day (dynamic range 120dB, THD+N -110dB), and still seems to be working reasonably well, though it's clearly no AP.
This shows the output at 1.998Vrms (since that's what Amir uses for his dashboard) without any load other than the 10kohm input impedance of the 1616m. Obviously this isn't going to win any prizes. The unregulated high-voltage and heater supplies are letting through mains ripple which shows up at -93dB @50Hz. While all connections are single-ended, the mains signal pollution through my Heresy is over 10dB lower, so this is really just a problem with PSU filtering, but it's good enough that I have never heard any trace of hum. I have no idea what's causing the strange noise hump at 7kHz, but at -116dB it's not a great concern. There's a long tail of harmonics, with the highest at around -85dB.
Here is the result from RMAA when unloaded, with the Heresy for comparison. I include the Heresy results to establish a baseline of the 1616m's accuracy.
The frequency response shows a mild (~0.6dB) roll-off at both ends of the spectrum. There is also a 0.6dB imbalance between channels, possibly because I didn't get a matched set of valves when I replaced them.
I was most interested in seeing how it performed under load, particularly as I use low-impedance headphones. Here are the power graphs at 997Hz using REW's new Level Step function into 19ohms and 314 ohms (the odd numbers are a result of me only having 0.25W resistors, so I paralleled a bunch of them together to fabricate the test loads). Voltages were calibrated using a fairly cheap RMS multimeter on a -6dB 120Hz signal which manages an acceptable amount of precision, but I can't guarantee the accuracy. Ignore the crosstalk figures in the RMAA results, as I only had one channel connected for these tests.
This is clearly happier with high impedances, being able to produce a healthy 170mW into 314ohms versus only 128mW into 19ohms. Distortion appears in three phases: first a ramp up from the noise floor followed by a plateau phase and then finally a sharp climb as the device saturates and goes into clipping. At the higher impedance the initial rise is steeper and the plateau phase lasts longer. I was intrigued to notice what appear to be periodic dips in the levels of the higher harmonics, so decided to investigate further at a range of frequencies in 0.5dB steps. These are animated gifs, which seem the clearest way to present these 3-dimensional data.
Here we can see that the dips, or 'distortion flicker' appears to be quite a robust effect. These can span a range of over 30dB, especially at higher frequencies, going from -80 to -110dB. While changing the load impedance alters the absolute levels at which they occur, the relative ordering of the dips appears unchanged. In order to be sure this wasn't some bizarre processing effect, I took screenshots of the RTA spectrum as the output level was moved across one of these dips and you can clearly see the 4th and 7th harmonics flickering down and then back up in line with the plot on the graph.
It's still possible that this was just some odd effect of the DAC and ADC in the 1616m, so I hooked up my old MacbookPro and ran a sweep using a TempoTec Sonata for output and the built-in line input for the ADC. This is unsurprisingly a lot noisier, but you can see the flicker effect happening to the same harmonics at roughly the same output levels. The only common factors here were the use of REW and the short cable I used to attach the test load.
With the output level held constant and distortion measured by stepping the frequency in 1/3rd octave increments we can see that the harmonics show a steady rise into higher frequencies with constant slope, apart from the 3rd harmonic, which has a negative slope in the sub-bass region.
Here are plots of the IMD and multi-tone responses included for the sake of completeness.
Sound
"But, but," I can hear the 'trust your ears' crowd asking, "how does it sound?" To be honest, for many months I had difficulty noticing any real difference between the XCans and my Heresy, apart from the fact that turning up the volume to ear-splitting levels (far higher than used for usual listening) would result in clipping. I've never attempted a proper double-blind test, largely because I don't have a couple of assistants willing to put up with the hours of tedium it would involve, so all comparisons were sighted. I believe that subjective impressions are strongly coloured by alterations in mood and mental state, and it can be difficult if not impossible to separate these from a genuine effect of the equipment.
But after living with this for a year I get the impression that bass through the XCans feels more 'lush', the rolling bass lines from LCD Soundsystem feel a little fuller and more satisfying through the XCans than the Heresy. On the flipside, the Heresy gives a greater impression of clarity and precision that's most apparent on complex classical music. Even though the Heresy is, from an objective standpoint, several orders of magnitude superior it must be stressed that the subjective differences are quite subtle. It's tempting to try to draw parallels between these subjective impressions and the rise in harmonic distortion at higher frequencies, but that would be speculation.
The most interesting result here is the distortion flicker effect shown above. Although the result from my MacbookPro goes some way to providing a control, this would have to be replicated in an entirely different setup to be certain it is a genuine effect. If it is genuine it seems unlikely that this is a truly novel finding, but my reading in Electronic Engineering doesn't go much further than a few basic textbooks. So it may well have been described before, but just not have received much attention.
If distortion flicker is genuine, then it has implications for attempts to replicate 'tube sound' using DSP. There are several plugins available that attempt to replicate valve amplifiers, and most of them seem to simply add in a fixed pattern of harmonics that's invariant with level. This also seems true of Paul K.'s Distort program, though it does show accelerated roll-off of higher harmonics as the level declines. The results shown here suggest that modelling these amps may require a more dynamic approach.
There are several examples of internet tests showing that most people have difficulty discerning added distortion even at quite high levels. Our sensory systems in general are much better at detecting changes in stimuli, as shown in this celebrated example from astronomy. It's certainly tempting to speculate that similar flicker effects in the level of distortion components may be detectable at far lower amounts than when the distortion level is static and thus easier to correct by our auditory cortex.
At the time the XCans was the subject of a bit of a modding fad, and you can still find a site selling capacitor kits here. As can be seen from the pictures, I tried my hand at this, switching the electrolytics for fancier 'audio' grade versions, bypassing them with polystyrene caps and changing the ceramic feedback compensation caps for polystyrene. Oddly enough, I couldn't hear the slightest difference after making this 'upgrade' ... When I exhumed the device from storage last year I noticed that one of the capacitors in the power supply was bulging, so replaced them all with 105C Panasonics. I added clip-on heatsinks to the BD139s (TR102) as there were signs of heat-stress on the PCB and drilled some holes in the casing to allow a small amount of airflow (as sold there was no ventilation at all).
The amp's transformer is in a wall-wart which feeds it 12VAC via a barrel connector. The valves are run as cathode-followers with an active transistor load and an opamp gain stage. It seems to be a unique hybrid design and I can't find anything similar in the resources on valve amplifier design. It's certainly very different to a White cathode-follower, which is the usual starting point for active-load CFs. One flaw that I found while tinkering with it is that if the valves aren't conducting it places 30VDC across the output. This also happens when it's powered up, as the PSU capacitors charge up before the heaters have had time to raise the cathodes to the correct temperature. So I added a timer relay to the input to the main diode bridge which turns it on 40 seconds after the heaters have received power. Fixing this properly would require a thorough redesign, but this works as a stopgap.
The output impedance is 49ohms, largely as a result of R115, the 47ohm current-limiting resistor on the output. When it was introduced it was championed as a good match for 300ohm Sennheiser HD600s, but I've found it works quite well with my ultra-low impedance (18ohm) planars, though this may be because they don't present a reactive load. Obviously with low-impedance headphones the majority of the power produced is actually being dissipated in the output resistor.
Measurements
I recently resurrected my 15-yr-old EMU 1616m audio interface running off an antique laptop with the requisite PCMCIA slot. While old, this boasted class-leading specs back in the day (dynamic range 120dB, THD+N -110dB), and still seems to be working reasonably well, though it's clearly no AP.
This shows the output at 1.998Vrms (since that's what Amir uses for his dashboard) without any load other than the 10kohm input impedance of the 1616m. Obviously this isn't going to win any prizes. The unregulated high-voltage and heater supplies are letting through mains ripple which shows up at -93dB @50Hz. While all connections are single-ended, the mains signal pollution through my Heresy is over 10dB lower, so this is really just a problem with PSU filtering, but it's good enough that I have never heard any trace of hum. I have no idea what's causing the strange noise hump at 7kHz, but at -116dB it's not a great concern. There's a long tail of harmonics, with the highest at around -85dB.
Here is the result from RMAA when unloaded, with the Heresy for comparison. I include the Heresy results to establish a baseline of the 1616m's accuracy.
The frequency response shows a mild (~0.6dB) roll-off at both ends of the spectrum. There is also a 0.6dB imbalance between channels, possibly because I didn't get a matched set of valves when I replaced them.
I was most interested in seeing how it performed under load, particularly as I use low-impedance headphones. Here are the power graphs at 997Hz using REW's new Level Step function into 19ohms and 314 ohms (the odd numbers are a result of me only having 0.25W resistors, so I paralleled a bunch of them together to fabricate the test loads). Voltages were calibrated using a fairly cheap RMS multimeter on a -6dB 120Hz signal which manages an acceptable amount of precision, but I can't guarantee the accuracy. Ignore the crosstalk figures in the RMAA results, as I only had one channel connected for these tests.
This is clearly happier with high impedances, being able to produce a healthy 170mW into 314ohms versus only 128mW into 19ohms. Distortion appears in three phases: first a ramp up from the noise floor followed by a plateau phase and then finally a sharp climb as the device saturates and goes into clipping. At the higher impedance the initial rise is steeper and the plateau phase lasts longer. I was intrigued to notice what appear to be periodic dips in the levels of the higher harmonics, so decided to investigate further at a range of frequencies in 0.5dB steps. These are animated gifs, which seem the clearest way to present these 3-dimensional data.
Here we can see that the dips, or 'distortion flicker' appears to be quite a robust effect. These can span a range of over 30dB, especially at higher frequencies, going from -80 to -110dB. While changing the load impedance alters the absolute levels at which they occur, the relative ordering of the dips appears unchanged. In order to be sure this wasn't some bizarre processing effect, I took screenshots of the RTA spectrum as the output level was moved across one of these dips and you can clearly see the 4th and 7th harmonics flickering down and then back up in line with the plot on the graph.
It's still possible that this was just some odd effect of the DAC and ADC in the 1616m, so I hooked up my old MacbookPro and ran a sweep using a TempoTec Sonata for output and the built-in line input for the ADC. This is unsurprisingly a lot noisier, but you can see the flicker effect happening to the same harmonics at roughly the same output levels. The only common factors here were the use of REW and the short cable I used to attach the test load.
With the output level held constant and distortion measured by stepping the frequency in 1/3rd octave increments we can see that the harmonics show a steady rise into higher frequencies with constant slope, apart from the 3rd harmonic, which has a negative slope in the sub-bass region.
Here are plots of the IMD and multi-tone responses included for the sake of completeness.
Sound
"But, but," I can hear the 'trust your ears' crowd asking, "how does it sound?" To be honest, for many months I had difficulty noticing any real difference between the XCans and my Heresy, apart from the fact that turning up the volume to ear-splitting levels (far higher than used for usual listening) would result in clipping. I've never attempted a proper double-blind test, largely because I don't have a couple of assistants willing to put up with the hours of tedium it would involve, so all comparisons were sighted. I believe that subjective impressions are strongly coloured by alterations in mood and mental state, and it can be difficult if not impossible to separate these from a genuine effect of the equipment.
But after living with this for a year I get the impression that bass through the XCans feels more 'lush', the rolling bass lines from LCD Soundsystem feel a little fuller and more satisfying through the XCans than the Heresy. On the flipside, the Heresy gives a greater impression of clarity and precision that's most apparent on complex classical music. Even though the Heresy is, from an objective standpoint, several orders of magnitude superior it must be stressed that the subjective differences are quite subtle. It's tempting to try to draw parallels between these subjective impressions and the rise in harmonic distortion at higher frequencies, but that would be speculation.
The most interesting result here is the distortion flicker effect shown above. Although the result from my MacbookPro goes some way to providing a control, this would have to be replicated in an entirely different setup to be certain it is a genuine effect. If it is genuine it seems unlikely that this is a truly novel finding, but my reading in Electronic Engineering doesn't go much further than a few basic textbooks. So it may well have been described before, but just not have received much attention.
If distortion flicker is genuine, then it has implications for attempts to replicate 'tube sound' using DSP. There are several plugins available that attempt to replicate valve amplifiers, and most of them seem to simply add in a fixed pattern of harmonics that's invariant with level. This also seems true of Paul K.'s Distort program, though it does show accelerated roll-off of higher harmonics as the level declines. The results shown here suggest that modelling these amps may require a more dynamic approach.
There are several examples of internet tests showing that most people have difficulty discerning added distortion even at quite high levels. Our sensory systems in general are much better at detecting changes in stimuli, as shown in this celebrated example from astronomy. It's certainly tempting to speculate that similar flicker effects in the level of distortion components may be detectable at far lower amounts than when the distortion level is static and thus easier to correct by our auditory cortex.
Attachments
Last edited: