Review of the Sapphire 4
I'll do this in two parts, first now and then add a coda after I put the Sapphire 3 boards back and compare again.
The defining characteristic of the Sapphire 4 is a slight softness of the treble. These is as a tendency to drop hard consonants, so "crack" sounds like "thwack" and "crash" becomes "whash". Brassiness is de-emphasized, woodiness is emphasized. Transients de-emphasized, textural and ambient details emphasized.
It seems reasonable to me to ascribe what I'm hearing to the distortion behavior, which is mainly 2nd harmonic and more prominent at higher frequencies and higher signal levels.
It's very much opposite character to the Sapphire 3, which was all about transients, transparency, and control. The Sapphire 4 is big and woolly and soft and cuddly. It definitely puts a stamp on the reproduction, but it is not all unpleasant. Damping out some of the treble energy makes a lot of recordings/sources sound better, and indeed the soundstage and overall intelligibility are superb.
How much I liked the Sapphire 4 depended very much on what I was listening to, and what I was listening through. The Phonoclone sounded pretty decent, the CrystalFET sounded very, very damp, the Asus STX soundcard overall sounded excellent. Classic rock sounded amazing, classical and chamber music a little bit tired and indistinct. Jazz an pop was a mix of good and bad.
Ultimately I expect I'm going to prefer the Sapphire 3 / Sonic Imagery Labs 995FET modules because at heart I'm a dynamic range/resolution/accuracy guy above all else. We'll find out shortly if this is true.
I'll do this in two parts, first now and then add a coda after I put the Sapphire 3 boards back and compare again.
The defining characteristic of the Sapphire 4 is a slight softness of the treble. These is as a tendency to drop hard consonants, so "crack" sounds like "thwack" and "crash" becomes "whash". Brassiness is de-emphasized, woodiness is emphasized. Transients de-emphasized, textural and ambient details emphasized.
It seems reasonable to me to ascribe what I'm hearing to the distortion behavior, which is mainly 2nd harmonic and more prominent at higher frequencies and higher signal levels.
It's very much opposite character to the Sapphire 3, which was all about transients, transparency, and control. The Sapphire 4 is big and woolly and soft and cuddly. It definitely puts a stamp on the reproduction, but it is not all unpleasant. Damping out some of the treble energy makes a lot of recordings/sources sound better, and indeed the soundstage and overall intelligibility are superb.
How much I liked the Sapphire 4 depended very much on what I was listening to, and what I was listening through. The Phonoclone sounded pretty decent, the CrystalFET sounded very, very damp, the Asus STX soundcard overall sounded excellent. Classic rock sounded amazing, classical and chamber music a little bit tired and indistinct. Jazz an pop was a mix of good and bad.
Ultimately I expect I'm going to prefer the Sapphire 3 / Sonic Imagery Labs 995FET modules because at heart I'm a dynamic range/resolution/accuracy guy above all else. We'll find out shortly if this is true.
Review of the Sapphire 4 : Coda
Back on '3'. (2520 FET modules)
First impression is of a very dry and very very remote sound. What Stan said about martinis, well there's a reason people drink them after all. I really missed that lush, wraparound, technicolor soundscape that plops you down in the middle of the action.
But crank up the volume a bit and the soundstage stretches out L-R, and the immense available reserves of low-distortion drive start to make themselves felt. Transients have just the right amount of curl, while the balance of ambient/foreground and brass/wood is more realistic. The overall presentation is fast and bright. At it's worst, stridently so (most likely originating upstream). The space between instruments is not as well delineated, the stereo image is more jumbled. For all that though I prefer the '3' right now as having what I feel is the more accurate tone overall.
****
The two boards have the same Z-reg regulation, and nearly the same output stage. The volume control, power supply are all the same, I just swapped out boards. The changes are limited to the voltage amplification.
Back on '3'. (2520 FET modules)
First impression is of a very dry and very very remote sound. What Stan said about martinis, well there's a reason people drink them after all. I really missed that lush, wraparound, technicolor soundscape that plops you down in the middle of the action.
But crank up the volume a bit and the soundstage stretches out L-R, and the immense available reserves of low-distortion drive start to make themselves felt. Transients have just the right amount of curl, while the balance of ambient/foreground and brass/wood is more realistic. The overall presentation is fast and bright. At it's worst, stridently so (most likely originating upstream). The space between instruments is not as well delineated, the stereo image is more jumbled. For all that though I prefer the '3' right now as having what I feel is the more accurate tone overall.
****
The two boards have the same Z-reg regulation, and nearly the same output stage. The volume control, power supply are all the same, I just swapped out boards. The changes are limited to the voltage amplification.
BOM 4.0b6
Recommended R2/R3 gain set resistors changed to the following. This cranks up the feedback screw, lowering distortion and drawing closer to the stability limit. I was overly conservative in my initial values.
R2 R3 gain
221R 10k 27 dB
330R 10k 24 dB
475R 10k 21 dB
681R 10k 18 dB
1k 10k 15 dB
1k5 10k 12.5 dB
2k21 10k 10 dB
4k75 10k 6 dB
10k 10k 3 dB
The BOM also included the feedback connection mod to the output, and suggested values R2/R3 to use there. They are chosen such that the circuit should still be stable without C2, but 47 pF compensation or so can be added is overshoot is detected. With the feedback mod, a Zobel filter on the output (47 ohms and 4.7 nF in series, connected output to ground) may also be needed to keep it all well-behaved.
Recommended R2/R3 gain set resistors changed to the following. This cranks up the feedback screw, lowering distortion and drawing closer to the stability limit. I was overly conservative in my initial values.
R2 R3 gain
221R 10k 27 dB
330R 10k 24 dB
475R 10k 21 dB
681R 10k 18 dB
1k 10k 15 dB
1k5 10k 12.5 dB
2k21 10k 10 dB
4k75 10k 6 dB
10k 10k 3 dB
The BOM also included the feedback connection mod to the output, and suggested values R2/R3 to use there. They are chosen such that the circuit should still be stable without C2, but 47 pF compensation or so can be added is overshoot is detected. With the feedback mod, a Zobel filter on the output (47 ohms and 4.7 nF in series, connected output to ground) may also be needed to keep it all well-behaved.
yes its me.. lol. been busy as of late so burnt fingers have been moved to the side for now.
Richard. by the sound of it those Sapphire 4 boards would suit an RPi perfectly. the RPi tends to be a bit brittle sounding so might get a bit tamed with the new Sapphire.
can I just ask what your conclusions were regarding 'in head' soundscape e.g. was the L/R taken out the headphones and placed more middle of head or visa versa?
keep up the good work
Richard. by the sound of it those Sapphire 4 boards would suit an RPi perfectly. the RPi tends to be a bit brittle sounding so might get a bit tamed with the new Sapphire.
can I just ask what your conclusions were regarding 'in head' soundscape e.g. was the L/R taken out the headphones and placed more middle of head or visa versa?
keep up the good work
@Bibio
"was the L/R taken out the headphones and placed more middle of head or visa versa?"
Not quite following you there. '3 gives you a balcony seat, music front middle with some LR separation that seems to widen with volume. '4 is more 3D wraparound and you feel an ambient presence - perhaps artificially so but the illusion is believable. Music in VR if you will.
"was the L/R taken out the headphones and placed more middle of head or visa versa?"
Not quite following you there. '3 gives you a balcony seat, music front middle with some LR separation that seems to widen with volume. '4 is more 3D wraparound and you feel an ambient presence - perhaps artificially so but the illusion is believable. Music in VR if you will.
Slowly but surely...
Thanks to the miracle of parallel resistors, my R2 measures 297.8 on one board and 297.6 on the other. R3 measures 12.7K and 12.71K, respectively. These values give me a gain of 26.73 dB and 26.75 dB, close enough (headphones 250 ohm, 96 dB sensitivity). Had to hunt through my stash to find values that would work for the new gain set. And I took seriously your comment about instability below R2=300 ohms, Richard.
Measuring and matching every component, board-to-board. Enjoying it, but my feet are getting cold working in the basement.
Thanks to the miracle of parallel resistors, my R2 measures 297.8 on one board and 297.6 on the other. R3 measures 12.7K and 12.71K, respectively. These values give me a gain of 26.73 dB and 26.75 dB, close enough (headphones 250 ohm, 96 dB sensitivity). Had to hunt through my stash to find values that would work for the new gain set. And I took seriously your comment about instability below R2=300 ohms, Richard.
Measuring and matching every component, board-to-board. Enjoying it, but my feet are getting cold working in the basement.
This is, indeed, enticing. I can't wait!
One thing I'm a bit concerned about is that my C1 on Sapphire 3 are Clarity Caps that aren't going to leave those boards nicely. The durned things were expensive, so I'm not buying them again unless I have to. I have a pair of .68uF Mundorf M Caps (ZN, I think) that I bought some time ago to try but haven't yet, so they will go into Sapphire 4. I won't have any way to distinguish the sound of the caps...
I recommend standoffs and/or keeping the leads of "precious" coupling caps longer than needed to fit the board. It's not as neat but makes refitting easier.
****
Re R2. I don't know exactly how low is too low. It depends on parasitic capacitance at the input for one thing. As long as the closed loop gain is large however (>20 dB) small values of R2 should not be a problem. As per the 40b6 BOM even 221 ohms should be ok at 27 dB gain.
All the simulation tells us is where it will definitely be unstable. There's a grey zone in front of that where actual testing would be needed to find out for sure.
****
Re R2. I don't know exactly how low is too low. It depends on parasitic capacitance at the input for one thing. As long as the closed loop gain is large however (>20 dB) small values of R2 should not be a problem. As per the 40b6 BOM even 221 ohms should be ok at 27 dB gain.
All the simulation tells us is where it will definitely be unstable. There's a grey zone in front of that where actual testing would be needed to find out for sure.
HFE scatter
I bought a hundred of each of the BC327 and BC337 transistors, expecting to measure and at least be consistent. The NPN BC337 were remarkably consistent in HFE current gain, all in the 250 to 270 range. The PNP BC327 were all over the map, with HFE ranging from about 210 to over 370, and not an even distribution, either. Quite wacky, but there were three bands in which there were sufficient to populate the boards with like values.
Just the TO-26 trannies to do, and the terminal blocks - tomorrow.
I bought a hundred of each of the BC327 and BC337 transistors, expecting to measure and at least be consistent. The NPN BC337 were remarkably consistent in HFE current gain, all in the 250 to 270 range. The PNP BC327 were all over the map, with HFE ranging from about 210 to over 370, and not an even distribution, either. Quite wacky, but there were three bands in which there were sufficient to populate the boards with like values.
Just the TO-26 trannies to do, and the terminal blocks - tomorrow.
I've had similar experiences, both with TO-92 and TO-126 parts. Normally though, my experience is that the PNP parts are consistent and the NPN parts vary wildly - not sure why.
/U.
Component value spread
I matched the Zeners, too.
Back to those transistors with some more info on the HFE spread in the PNP BC327:
I separated them in value bins of ten. Even without graphing the data, it looks like a messy trinomial distribution:
18% were in the HFE 210 & 220 bins
none between 220 and 280!
1% in the 280 bin
22% in the 290 bin
19% in the 300 bin
8% in the 310 bin
14% in the 320 bin
6% in the 330 bin
3% in the 340 bin
10% in 350 to 370 bins
It does re-enforce the notion that one ought to buy more transistors than you think you need!
I matched the Zeners, too.
Back to those transistors with some more info on the HFE spread in the PNP BC327:
I separated them in value bins of ten. Even without graphing the data, it looks like a messy trinomial distribution:
18% were in the HFE 210 & 220 bins
none between 220 and 280!
1% in the 280 bin
22% in the 290 bin
19% in the 300 bin
8% in the 310 bin
14% in the 320 bin
6% in the 330 bin
3% in the 340 bin
10% in 350 to 370 bins
It does re-enforce the notion that one ought to buy more transistors than you think you need!
Are yours the Fairchild parts as per the BOM, from a reputable source like Mouser? I've never encountered anything like that before, and I've measured 100 unit batches on multiple occasions. I would say 90% of the BC327 would be 230-300, with <5% 300+ or under 200.
The BC327-25 is supposed to have a spread of hfe 160-400. With nothing in the range 220-280 your parts look suspiciously like B-grade leftovers someone else already cherry-picked from.
The BC327-25 is supposed to have a spread of hfe 160-400. With nothing in the range 220-280 your parts look suspiciously like B-grade leftovers someone else already cherry-picked from.
Last edited:
Well they are all technically within the 160-400 values the grade specifies them to be so while odd if they came from Mouser there is nothing suspicious. Someone at some point, maybe Fairchild themselves, binned them that way. Maybe needed some matched ones for a big customer, sold the rest on to Mouser.
****
BOM 40b7 uploaded. The heatsink part number on the previous version was incorrect, it was 532-507302B00 but it should be 532-577002B00 (similar but narrower model).
Thanks to Stan for finding that error.
****
BOM 40b7 uploaded. The heatsink part number on the previous version was incorrect, it was 532-507302B00 but it should be 532-577002B00 (similar but narrower model).
Thanks to Stan for finding that error.