🚀 go-pugleaf

   Not being content with the way things are, and being an audio
junky, it wasn't long before my poor (almost new) LINN LK100
succumbed to tweak fever. I was pretty impressed with it at the
hi-fi store, but then a pretty nasty review from 'HI-FI News and
Record Review' (May 1992) planted discontent in my head.
Don't you just hate to see a negative review of the thing you
just bought, those golden ears you thought you had, are just
gold leaf covered tin ones....

	Things I liked about the Linn LK100:
1: It's reasonably compact, doesn't weigh a ton, puts out about 50 W
   per channel into 4 ohms.
2: Sounded pretty good at the stereo store, beat out the NAD amps,
   and was indistinguishable from the LK280 (about $2500).
3: Is really easy to strip apart, I can remove ALL the main components in
   about 5 minutes. The circuit board is nicely laid out, with the
   names of all the parts silk-screened on the fibreglass board.
4: Plenty of outputs (banana plug type), 2 sets per channel. It would
   have been nice if they had spaced them properly, so you could use
   a dual banana plug. That's British for you.

	Needless to say, I didn't lift the lid for about 3 months,
and then my curiosity got the better of me. A number of things met
my eyes that I didn't like, so I spent a few evenings and made up
a schematic for it. In doing that I found a number of other things
that were rather detestable. Finally, I put a signal generator and 'scope
on the unit and made a few measurements, some good, some bad.
	Let me describe the circuitry. It is a fairly standard
audio amplifier. First there is a differential pair, emitters driven by
a current source, and active collector loads for high gain. The one
side of the differential pair is driven by the audio input, the other
side by the feedback network from the amp's output. The output of the
differential amp goes to a 1 transistor stage whose collector drives
the amplified diode biasing circuit, and again, an active load.
The amplified diode sits across the bases of two complementry medium
power transistors, whose emitters are joined (through diodes) to the
speaker load. The collectors of the medium power transistors drive the
bases of the main output transistors (Sanken type). The output
transistors are run common emitter, their collectors are connected
together (through diodes) to the speaker load. This is the only
slightly novel thing about the circuit topology.
	The designers that laid out the board were a little more
innovative, with the exception of the power transistors, all the 
rest are surface mount, one kind of NPN, one kind of PNP. The
medium power types are a composite of three small ones. The transistors
were the only surface mount components (whew!).

	Well, since I didn't change any of the good things, I'll
list them first.
Good things I found when playing with the LK100's insides:

1: Nice heavy toroidal power transformer.
2: very little noise. I can just barely hear it with my ears up to
   the speakers.
3: Good high frequency response, -3db at about 50 kHz or so.
4: Low distortion, I didn't measure it at the time, but they claim
   somewhere below .02% .

Here's the bad things, and what I did to improve them. At the end, I'll
write a summary of what audible effects they had.

1:  Supply electrolytics were too small, at least so I thought. They
   were 10,000 mfd, 50vdc. I added another 10,000 to each rail, and used
   Mallory LP series, which are supposed to have a pretty low ESR. In
   addition I added some 1 mfd. very low ESR film types. The wiring
   was done in such a way as not to disturb the ground paths. The
   LINN has multiple ground lines, for the raw power supply, for the
   signal and feedback circuits, for the regulated low-power supplies,
   and for the speaker returns. Any change in wiring to these lines,
   may result in all sorts of horrible behaviour , so I was careful
   to add the extra capacitors in parallel with the terminals of the
   old ones. 
2:  The emitters of the output transistors didn't go directly to the
   rails. Instead, they joined the other emitter lead from the other
   channel and THEN went through 0.1 ohm to the rail. Where they joined
   together was bypassed through a cruddy little electrolytic (100 mfd)
   to ONE of the speaker returns (ground). This little trick puzzled
   me for a long time, I still don't know what they were trying to
   accomplish. My best guess is it's either a fault current limiter,
   or an attempt to decouple the raw supply. What I think it does
   is cause crosstalk between channels. So what I did was give each
   emitter its own 0.1 ohm resistor to the rails, and bypass each
   one through a very high quality 470 mfd. electrolytic and a 1 mfd.
   polypropylene capacitor (one meant for filtering high frequency
   switching supplies). The grounding I'll talk about later....
3:  On overload, the amplifier oscillates at about 500kHz, but only
   during the overload condition. Of course it raised hell with the
   other channel because of the LINN's poor crosstalk. This was cured
   by adding .003 mfd. between collector and base of the amplified
   diode transistor. This didn't affect anything during non-overload
   conditions. (on the circuit board, this was effectively in parallel
   with r309 and r209)
4:  The output transistors' bias currents seemed a tad low (80 ma.) so
   I ran it up to 150 ma. by putting 2000 ohms in parallel with r210,
   (a 220 ohm resistor between base and emitter of the 
   amplified diode transistor). This, of course made the transistors
   run much hotter, so I opened up the grillwork of the case, so that
   more air could flow past the heatsinks inside. This can be done
   without making a mess of the case, and without any extra risk of
   having things or fingers get into the circuitry. The standard test
   for louvres and openings will still pass, at least according to
   our electrical safety codes (CSA). The temperature at the heatsink
   is between warm and hot, but not painful.
5:  The feedback network uses tantalums in the audio path (yechh). I 
   don't have that much against tantalums, from my own measurements they
   are somewhat better than aluminum electrolytics (the common kind) at
   high frequencies, ie the ESR stays fairly low up to 50-100 kHz.
   What they don't like is to be run with a DC voltage less than half
   their rated voltage. When this happens, they usually start to change
   value, and become unreliable (they usually short out). The feedback
   network consisted of 3.9kohms between the output and input to the
   one side of the differential pair (the inverting input). There was
   150 ohms in series with 120 mfd (tantalum) going to ground. With
   full output, I couldn't see any trace of signal (above 1 kHz) across
   the tantalums, so they didn't affect the high frequencies as I might
   of expected. Not leaving well enough alone, I changed the network.
   I multiplied the two resistors by 4, thus requiring 1/4 the value of
   capacitance. Instead of tantalums, I used 40 mfd's worth of film
   capacitors (polyester I think, polypropylenes are just too big and
   expensive). Well I thought there was an improvement in bass, so
   I assumed that the tantalums had drifted off from their nominal values.
   After just measuring their values a few minutes ago (I saved all the
   old components) they were only 5% less than nominal. So much for
   hearing tests. I suppose the designers figured that even with
   so little DC across them, there was even less AC, so it's just like
   they were sitting on the shelf unused. I had talked about the
   tantalum issue to a couple of LINN rep's at an audio show, they
   had heard about the arguments pro and con, and the word went out
   to stick with tantalums. Oh well....
6:  There was no line filter. Not a really major issue, since I use
   one of the no-name line filter boxes with EMI suppression. The
   Linn rep's told me that they could not get CSA approval on the ones
   they had designed in. I suspect that the reason for this is that
   the LK100 depends on the 15 amp fuse in your distribution panel for
   protection. From the point of view of an audio nut, this is a
   GOOD THING, since it means the power supply is effectively a very
   low impedance, in order to clear the mains fuse. The Linn supply
   as I saw it, does indeed have a healthy low impedance. I think the
   Linn people could not find a suitably cheap line filter that could
   withstand the 15 amps, maintain a low series impedance at 60Hz, and
   filter the grunge effectively. The best tweaks I could come up with
   were to put polypropylene caps across the primary and secondary of
   the power transformer. Not all capacitors can be used for this
   function! Only use CSA or U/L approved 'safety' or 'line' caps.
   A good choice is the Siemens "Y Safety" series, their number
   is B81121-C-B-14x, where x is a digit that specifies the capacitance.
   This capacitor is polypropylene, and can withstand 1500 vac, will
   not fail with up to 200V/us, and has a flame retardant construction. 
7:  There is a big power supply 'loop', in other words, the power supply
   rails and the returns are not close to each other. Under high 
   current, this acts like a loop antenna, and radiates a magnetic 
   field that induces voltages in other nearby wiring. Running
   a power test at 10kHz square wave, I couldn't see small signals on
   the board with a good oscilloscope probe, since they were all
   drowned out by the induced voltages in the measurement loop. Even
   when the probes' ground clip was connected to the input (it's only
   2-3 inches long) there was a definite signal.  The cross-talk
   test I used consisted of driving one channel full output at 10kHz
   square wave, the other had the input shorted. I then monitored
   the output of the 'quiet' channel. Both channels had a 5 ohm
   load. The so-called 'quiet' channel had about 800 mv. peak-to-peak
   hash, that ocurred on the square wave edges. This hash wasn't just
   a single spike as one would expect, but a 100-200kHz raggedly damped
   oscillation (maybe 5-10 cycles). 
     I rewired the power supply lines, speaker returns, decoupling
   returns, output lines and used #14 wire instead of the circuit board 
   tracing. It was a real pain to do. All wires that carried significant
   current were effectively paired with their return lines. In this way,
   the magnetic radiating cross-section was reduced by a factor of
   50-100. It also required considerable head-scratching, the optimum
   way to do this is not really intuitive, and will vary from layout
   to layout. 
     The rewiring dropped cross-talk from 800mv to about 80mv. Using
   heavy (#14) wire for returns, instead of the circuit board, dropped
   the cross-talk further to 20-30mv. Placing a 2700 pf from base to
   base of the output transistors reduced this to 10-15mv, and quieted
   down spikes that cross-talked during overload.
8:   Looking at the feedback circuit, I noticed their was no capacitor
   in parallel with Rf. (Rf goes between spkr o/p and base of differential
   input pair. Rg goes from that base through Cg to ground. Mid band
   gain would be 1+Rf/Rg.) Most of the amps I've seen usually have
   this capacitor which goes in parallel with Rf, which I assume gives
   an improved margin of stability.
     Anyhow, I figured out the value to start rolloff at 25-30kHz, and
   stuck it across Rf. It slowed down the response, removed a faint
   square wave overshoot (<3%), and further cut the crosstalk as
   described above. At this point crosstalk was about 5mv., way down
   from the original 800mv. If I call the added capacitor Cf, the 
   following are the new and old values.
     Rf(old)=3900 ohms     Rf(new)=15000
     Cf(old)= 0   pf       Cf(new)=150 pf
     Rg(old)=150  ohms     Rg(new)=604 ohms
     Cg(old)=120 mfd(tant) Cg(new)=40  mfd(film)
9:  Piddly little problem with the power LED - it's quite directional,
   you either can't see it, or it glares in your eyes. I stuck a small
   piece of translucent splicing tape in between the LED and it's mounting
   hole. No glare now, you can see it at almost any angle.

  And now for the REAL test, just what difference did these things do to the
sound? 
    1) beefing up supply caps  - no effect that I could tell, except the
       amp has about twice the 'run-on' time when it's turned off.
    2) decoupling o/p transistors - seemed to make an improvement in my
       ability to spacially locate things, although it might have been
       the coffee....
    3) killing oscillations during overload - couldn't here any change,
       I never have it up that loud.
    4) higher bias current  - made the sound less grainy, not a really
       drastic overall change though.
    5) removed tantalums from feedback path - couldn't hear any difference.
    6) putting line filtering in (includes plug-in type) - couldn't tell
       any difference.
    7 & 8) rewiring supply, adding feedback capacitor - made a real
       difference to the highs, especially violin music. MUCH smoother.
       Made several CD's that were unlistenable into acceptable sound.
       At first I thought I had miscalculated and dropped the top end
       down too far, but it was only about -1db at 20kHz.
        I'm beginning to suspect that some of the high frequency 'junk'
       that comes from CD players can really screw up the sound, and
       careful design of the high frequency response of an amp can
       reduce the audible impact of that 'junk' (through IM distortion,
       rectification of RF, etc. As a side note, I measured several
       millivolts of up to 100 MHz RF comming out of the audio lines
       of my CD player)

All in all, I feel the time I spent was worth it, it's probably doubled
the value of the amp, at least to my ears.
  I can FAX the schematics to those interested, of course I can't
guarantee their accuracy, or that you won't blow your amp up by fiddling
around with it.

-Paul

-- 
-----------------------------------------------------------------------------
Paul J Guy			work phone:519-885-1211 ext 6371  
paul@gaitlab1.uwaterloo.ca	home:519-744-5016   FAX:519-576-3090
pguy@healthy.uwaterloo.ca	..remember...bullshit baffles brains...