diyAudio (
-   Pass Labs (
-   -   1M-F5 - A 1 MHz F5 project (

triode_al 26th September 2011 07:07 PM

1M-F5 - A 1 MHz F5 project
3 Attachment(s)
My F5 project is almost down to finishing, I used 2SK175/2SJ55, one set.
The first measurements are very seducing:
Bandwidth > 1.000 kHz at 5 V RMS. Go think that.

Square wave: both channels have just the slightest and well damped overshoot, loaded on 4 or 8 ohm . Example: Square wave, 100 kHz, 4 ohm NI load
Attachment 241899

Square ware, unloaded
Attachment 241900
In this second picture you can see spurious overshoot that has some UHF components, maybe due to the long lead wires.

Photo 900 kHz on 4 ohm non-inductive load, at -3 dB
Attachment 241901

This is an incredible result, where 0 dB is 3,3 V RMS. And the board is underneath the chassis, the FETS are attached with 15 cm wires to the massive heat sinks on the side of the cabinet. Anyway, there is no sign of slew rate limiting, it is still a recognizable sine. But a square is not possible at that frequency.

Before connecting the output I measured the bandwidth of just the input: about 400 kHz at -3dB. [I test every component and half circuit as if it is an independent module].

triode_al 26th September 2011 07:09 PM

Some measurements on this 1M-F5
1 Attachment(s)
Power output and distortion
DC current: 1,8 amp, power rails 20 V / 20 V.
Max power, set to when slight clipping sets in on scope.
2 ohms, 7 V RMS, 25 watt
4 ohms, 9 volt RMS 20 watt
8 ohms, 11 V, 15 watts
As you see, this does not qualify as a burning amp. But with 175 Watt idle from the mains, it does

At 8 ohm, I get a damping factor of 20.
I found this formula: DF = V1 / (V2-V1)
V1 = output voltage of the amplifier with load resistor connected (8,71V)
V2 = output voltage of amplifier without a load (8,28 V).

Attachment 241903
foto clipping
Nice behaved.

Channel balance: -0,040 db unloaded, and loaded with 8 ohm: -0,063 db
I obtained this result by careful selecting the feedback pairs.

Nelson Pass 26th September 2011 10:00 PM

I seem to recall that a stock F5 does just about that with -3 dB at 1 MHz,
but I didn't record data that high, not seeing the point...


triode_al 27th September 2011 06:54 AM


Originally Posted by Nelson Pass (
I seem to recall that a stock F5 does just about that with -3 dB at 1 MHz,
but I didn't record data that high, not seeing the point...


I hope to surprise you even more, later :)

triode_al 27th September 2011 07:08 AM

Power supply
4 Attachment(s)
I made a full and complete star earth.

photo's of the starred earth
Attachment 242003 Attachment 242004

As diodes I used hefty diodes: 1N3891, fast recovery rectifier, 200 V, 12 A RMS continuous. They are mounted on a heat sink.

I have in first instance used a CLCRC power supply So the two rails have a different voltage (19,7 and 18,5 volt). The first cell has a 2.500 muF; then an inductor; the second cell 50.000 muF; with 0,22 ohm to the final cell 2*68.000 muF; with the scope I saw that the fluctuations from the power sector are very very low (a few mV only up and down). Nevertheless, bandwidth and power specs are even obtained with this lopsided power supply.
The inductors in the power supply are very effective. They are double-C. RMS hum reduces :
V-, raw 19,5Vdc; AC 1,9 volt in the first cell to 3,1 mV (-55 dB !!!!) in the second cell and 2 mV in the final cell ;
V+, raw 18,5 Vdc; AC 2,2 volt in the first cell to 6,6 mV in the second cell and 3 mV in the final cell.
Attachment 242006

These DC differences had to be sorted out - are the two inductors different or are the two initial capacitors different? But I bought the inductors cheap in a dump shop, so if they differ, can't buy others anymore. If it is the capacitor, then I can swap the caps.
I tested the caps, these are 2.600 and 2.700 muF respectively. OK. It was the two inductors that do have different specs (though look the same): 0,56 ohm and 1,12 ohm.

I changed to normal resistors and changed the first cell of the power supply from CLC to CRC type. With a 0,47 ohm resistor I get:
both V+ and V-, raw 19,5Vdc; AC 1,1 volt in the first cell to 65 mV (-25 dB only) in the second cell and 3,9 mV in the final cell ; final cell has 19,70 dc volt both halves.

Attachment 242007
The line perturbation can now be seen on the scope on the power supply but not on the output; the power supply noise is cancelled (which was not the case with the mis-matched inductors).
If I want to go sophisticated, I can make a shunt in the second cell. In my Le Monstre (an 8W design by Hiraga, in broad lines comparable with the F5) I have implemented a shunt power supply. This is very effective for the background details.

triode_al 27th September 2011 07:26 AM

3 Attachment(s)
A bit overdone, but I had these big heat sinks, and I planned to go this big capacitor route.

Attachment 242010
As you see, I have implemented an array of 4 pairs of FETS for each channel, I have the intent on paralleling.

Attachment 242009

Front side
Attachment 242008

Maybe ugly in proportion, specifically if you note it just is a 20 Watt Monster.

With this array, I might go BA-2 and make a passive output stage, using the F5 board just as a BA-3 driver (with reduced current of course).

triode_al 27th September 2011 07:30 AM

Parts and components (1)
2 Attachment(s)
I used the F5 boards from Jims audio store. Very well made, truly military grade. I like it that they are a bit bigger so leaves room for parts. Though some want to create a smaller board (Peter) with the shortest signal path, this board allowed me this bandwidth of 1+ MHz.
I have thought about the cascode (CViller in DIYshop) because the thermal load on the input will be constant (if you implement the cascade correctly), it will give slightly more 2nd harmonic, and some say the sound is more dynamic. By hey, look at a bandwidth of over 1 MHz - do I need more dynamics???

The 2SK170/2SJ74 input pairs. I bought matched input sets from AudioFet (Alweit) on ebay. These are marvelously paired. I tested them beforehand on both Idss @ 10 V and next on 20 V with source resistors of 0, 5, 10, 15, 20, 25 ohms.
I also tested the sets when I still had no output attached: equal across the both channels within 0,5 dB. I must say, Aleit did a very good job . It is not necessary at all to use different source resistors as EUVL suggested with these pairs.
How he packs them is shows his care.
Attachment 242012

Attachment 242013
specs of the inputs: impressive
Note that due to thermal drift, I cannot show best matching measurements, I doo it statically (not with pulse like a professional tester will do)

(Note: Previously I had bought unmatched jFETS elsewhere, but these measured all over the place while the 2SJ74's were fakes. So really go for a dependable source with matched pairs like AudioFet.)

As drain load for the input I choose 680 ohm carbon film (has low inductance) // 5 k 25 turns pot. For these transistors 560 ohms works OK too (for other FETS this resistor will be higher).

triode_al 27th September 2011 08:08 AM

Parts and components (2)
Output with 2SK175 / 2SJ55, in a big TO-3 housing.
My output FETS were bought on the local market in Guangzhou from a surplus refurbishing shop ( I bought 22 pairs, costed a 'small fortune'); they measure exceptionally well. The 2SK175 / 2SJ55 has slightly higher specs than the more commonly known 2SK135/2SJ50: a higher current, voltage and a higher power - 125 watt against 100 watt). They have been used in very high end audio amplifiers. The Ciss is 2SK175 = 700 pF, 2SJ55 = 1.000 pF, about half of the stock F5 outputs.
I have implemented four pairs per channel, but first only connected one pair.

# @100 mA @600mA delta, + 0,5A=
1 0,647 1,520 0,87
2 0,664 1,616 0,95
3 0,671 1,603 0,93
4 0,705 1,573 0,87
5 0,714 1,656 0,94
6 0,716 1,647 0,93
7 0,718 1,638 0,92
(Tested with a current source LM323 and a 50ohm/10 ohm resistor, fet in air, so does not consist of a good test, but adequate for matching; all readings after 5 seconds)

I flesh mounted them on the heatsinks, took me quite a time [days] with the precision drilling, cutting the screw mount, and ensuring there is no short-circuit between source and heatsink. You cannot drill this precision by hand.

Gate resistors: I soldered 100 ohms allan bradley pressed carbon resistors (these are non-inductive) on the gates; and inserted similar resistors in the boards, so both ends of the gate connecting wires have a resistance. I ensured the drain wire is physically routed separate from the gate wire.

Source resistors: I used 0,47 ohm Fukushima Futuba drains resistors (non-inductive, cemented, 6 W), selected, with the highest value for the 2SK175 and the lowest for the 2SJ55 (because the knee is slightly different, about 200 mV between the units). That will give a near equal driving voltage in the input.
They do get very hot to the touch at 1,5 - 1,8 amps dc, so bolting them to the chassis with a washer or giving them a heat sink is really mandatory for the amplifier to remain stable over the years. These resistors have pins that get loose very quick, be cautious when bending!
These non-inductive resistors are a key to the bandwidth.

Current limit: I did implement the current limiter, at approx. 6 amps (safe, very safe margin for this MOSFET that can handle 9 amp pulsed at 10 volt ds), by using 820 ohms/200 ohms divider network instead of 1k/150.

I have not used the thermistor - the 2SK175 / 2SJ55 have a very nice negative temperature compensation behavior. That is why I changed the drain resistors for the input pairs.

triode_al 27th September 2011 08:44 AM

Parts and components (3)
2 Attachment(s)
Current limit: I did implement the current limiter, at approx. 6 amps (safe, very safe margin for this MOSFET that can handle 9 amp pulsed at 10 volt ds), by using 820 ohms/200 ohms divider network instead of 1k/150.

I have not used the thermistor - the 2SK175 / 2SJ55 have a very nice negative temperature compensation behavior. That is also why I changed the drain resistors from 2.2 k to 680 ohm for the input drains.

I took very much care to select the feedback pairs, as I believe that a mismatched top/bottom half of the amplifier (they are both in parallel) will otherwise compete with each other, leading to oscillations. The top and bottom effectively are independent amplifiers, specifically so with the separate feedback loops of the F5.
Attachment 242026
See that it works out

First I bought 100 ohm 3W metal foil resistors (good within 0,1%, Sunpec) but later settled for 50 ohm carbon film because these allowed even tighter matching with the 10 ohm resistors that differed a bit more.
Attachment 242027

triode_al 27th September 2011 08:54 AM

Set up procedure
1 Attachment(s)
Before starting, I had tested what bias is needed on the outputs (about 1,6 V for 0,5 amp); from that I had selected an inpout drain resistance that gives a good amount of adjustment (680 ohm//5k trim pot).

Set all pots to 0 ohms. I used a 0,5 amp current source (LM323 with 10 ohms load) to initiate the current on the plus side (2SK170/2SJ55), then I inserted a 4 ohm resistor, and balanced the bottom side (2SJ74/2SK175), this final balancing needs quite some iterations, I initially choose 0,8 volt over the 0,47 ohm (1,8 amp). [Some might think this is too complicated]
Instead of a 0,5 amp current source it is just as easy to connect a four ohm resistor and set the top to 0,5 amp (2 volts). This is what I did in the second channel. And then start to increase current on the other side. The whole current then balances to about 1,5 amp. The process works oh so sweet. :)

Attachment 242034

A very nice and quick procedure; compare that with the following: I once spent a day swapping fixed resistors on the Le Monstre that shares the basic topology with the F5 but uses fixed resistors . . .
The DC balance is very good, it stays within mV of zero, except on heavy load, then it will change a some 20-30 mV.

When I will parallel the outputs (later, in due time) then I will first set the current to about 0,5 amps (120 mV), altogether then I will watch the total current.

All times are GMT. The time now is 03:21 AM.

Search Engine Optimisation provided by DragonByte SEO (Pro) - vBulletin Mods & Addons Copyright © 2018 DragonByte Technologies Ltd.
Resources saved on this page: MySQL 18.75%
vBulletin Optimisation provided by vB Optimise (Pro) - vBulletin Mods & Addons Copyright © 2018 DragonByte Technologies Ltd.
Copyright ©1999-2018 diyAudio