One of my clients recently asked me if he could get better music with a tubed preamp
as an alternative to his Sutherland  phono amp.
The Sutherland is all solid state using opamps, battery powered, and cost several thousand dollars.

I said yes, I could do better with tubes, but I would have to use just the ONE solid state device, a single 2SK369 j-fet as the input device.
I was then contracted to proove what I said.

My client does have enough quality elsewhere in his system so that the benefits of a better phono amp stage
would be discernible.
He has a nicely sized room which is well carpeted and furnished to reduce reverb and glare from too much treble. His speakers are Vienna Acoustic Mozarts, with Townsend super tweeters, and a custom built
subwoofer, ( also something I built )  and Musical Fidelity A3CR sub amp. The power amps are 22 watt mono bloc SEUL 13E1 that I built in 1997, upgraded for 2003, and the line stage preamp is also one of mine and which I also upgraded to the schematic shown below in 2005.
His vinyl replay uses a Michel Orb TT with an extensive sand and wire damping kit from a dude in
Denmark, the cartridge is Helicon Lyra.
Cabling is thin wire home brew plaited for interconnects, and Nordost speaker cables. So all up, this system is one of the few I am very happy to spend hours in front of, and despite my client's pair of good quality silver disc players for CD, SACD, DVD, even including a Sony player with Allen Wright's special upgrade.
 
We both find that vinyl sounds better, and is more emotionally engaging, unless the recording isn't good.
It is remarkable how often they got it right in the studio/venue in the analog 1970s, and they cannot
do much better now when surrounded by digital gadgets, and even with 96/24bits.

Perhaps the digital replay gear could still be better, but we can't afford the very best....

Gary's system I can enjoy.....

Directly below the Michel TT is a Marantz tuner, then the "Rocket" phono amp I built.
The line stage preamp and power supply for both line and phono stages are to the left of the
silver coloured Marantz tuner.

The "Rocket" six tube phono amp, dec 2005. It rocks!

Here is the ""Rocket"" schematic for the latest phono stage, 2005....

The basic layout is all single ended topology with cascode j-fet + triode input stage and
µ-follower output stage. Passive RIAA is all done in one network between stages.
I did at first try a fully balanced differential LTP input stage with conversion to
unbalanced output in a quasi balanced output stage but the complexity was much worse than what you see above and I had more trouble trying to keep hum and noise low. I have settled for all unbalanced topology
because really, who the hell really needs balanced? This circuit gives low noise, wide bandwidth,
low distortion, and oddles of music, and can be used to power any tubed or solid state preamp
or sound card for recording vinyl onto a CD. What more does anyone want?

The load for the cartridge can be altered to any value by soldering a chosen R across two
points on a board provided within the amp.  The gain can be altered by soldering a link on two points on
same board, so without the link, gain at 1kHz = 45dB, with the link, = 57dB, or 12dB more for MC.
So if MC rated output is as low as 0.2mV, output = 140mV. 

The j-fet input gate is biased at 0V and the source resistances R3 to R11 are taken to a shunt regulated negative voltage also held steady by 10,000 uF caps.
The set up allows for a minimum effect of electrolytic bypass caps in the j-fet source circuit,
and also no caps between the j-fet gate and input terminal.

The cascode circuit works as follows :- The input signal is applied to the very high input impedance
gate of the j-fet. Any input voltage signal across the gate to source, Vgs of the j-fet makes a current change through the drain to source path = gm x Vgs. This current change occurs also within the tube and the load
at the anode, and in the j-fet source resistance between source and 0V.

Let us consider that the source to 0V R = 30 ohms when the link for high gain is in place as shown.

The load seen by the V1 tube varies slightly from a maximum of 23.5k (ohms) at very low F to about
15k at 1kHz  due to the nature of the impedance of the RIAA filter which varies with F.

Let us consider the situation at 1 kHz. V1 tube gain, A,  = µ x RL / ( RL + Ra ).

In this case V1 paralleled 6DJ8 Ra = 2.5k approx, and µ = 33, so A = 33 x 15 / ( 15 +2. 5 ) =  28.2. approx.
( Ra varies considerably with the idle  anode current and for 1/2 a 6DJ8 Ra = 5k approximately
so for the paralleled tube it is about 2.5k; an approximate calculation is all that is needed to explain
how the circuit works. )

Consider V1 operation first. Since the V1 grid is effectively grounded, the V1 voltage gain depends on voltage change at the V1 cathode. V1 is operating in grounded or common grid mode, which is unusual, but very
effective in some situations where we wish to avoid the Miller effect. Unlike where we might otherwise alter the the input voltage of the high impedance input grid to cause an anode voltage change and tube current change, we must apply a change in voltage *and* current to the cathode when the grid is grounded, because the cathode is a conducting terminal in the circuit of tube, load, and power supply. Since we know voltage gain is 28.2, then the change of cathode voltage will be 1/28.2 of the change at the anode, but the current change will be exactly the same, so if we know RL = 15k then the impedance looking into the V1 cathode is 15k x 1/28.2 = 532 ohms.
This impedance is the load seen by the drain of the j-fet.
The  gm of the 2SK369 at 55.2mA of idle current = 40mA/V, and since fet voltage gain = gm x RL,
then in this case fet gain = 0.04 x 532 = 21.28.
But we also have 30 ohms between the source and 0V which acts to give some local current NFB
to the j-fet circuit, so that j-fet gain is reduced and gain with NFB = A / ( 1 + [ A x ß ] ),
where ß = the fraction of the output signal fed back in series with the input to gate.
In this case, ß = 30 / (532 + 30) = 0.0534, so gain with NFB = 21.28 / 1 + [ 21.28 x 0.0534 ] ) = 9.96.

The total gain between input gate and V1 anode drain = j-fet gain x tube gain = 9.96 x 28.2 = 281.0.

There is a simpler way to consider all this.
It can be proven that the total gain  for the j-fet and V1 is simply j-fet gm x tube RL.
The maths wizards amoung you will understand why from what i have said so far, and they will also know
that almost any tube can be substituted for V1 providing there is no grid current, so a trioded 6AU6, 6EJ7,
or triode such as 12AT7, 12AU7 can be used and the gain between input gate and tube anode will not change!

So if tube RL = 15k, gain ( without current feedback from the fet source R )  = 0.04 x 15,000 = 600.
With Rs = 30 ohms, ß = 30/15,000 = 0.002

Gain with current FB = 600 / ( 1 + [ 600 x 0.002 ] ) = 272.7, very close to what we calculated
before with a two stage gain calculation.

There is more to consider with this cascode circuit. The j-fet operates with a voltage gain of about 10 between input gate and drain. J-fets, like triodes, have considerable capacitance between the drain and gate,
so with any gain there is Miller input capacitance, but since the MC has such a low impedance
the Miller C will have a negligible effect on the response from the cartridge. But if a moving magnet cart
with much higher output impedance was connected the Miller C would need to be considered when working out the shunt C needed by the MM cart for the correct response for that MM cart, often about 360pF, and the C must allow for cable C, and the Miller C.
Fortunately, the j-fet gain is only 10, so the Miller C is very low; lower than a high gain input tube such as a 12AX7 with a gain of say 80.
Using the j-fet as a cascode driving device will always result in having a low voltage gain and low Miller effect.
If you didn't have the V1 tube sitting on top of the j-fet then the j-fet gain would be simply 600
 ( without current FB, ) and the Miller effect would be huge. V1 could be another j-fet and the source input impedance would be 15,000 / 600 = 25 ohms, so the bottom driving fet would have a voltage gain
of 0.04 x 25 = 1.0, and Miller C would indeed be low, and where we wanted really low MillerC
then we might use two j-fets in cascode. In this case we wish to amplify slow moving audio signals,
and we wish to have a huge dynamic range and low THD, so we make the top device of the cascode pair a tube.

The impedance looking into the drain of the j-fet is very high, perhaps 50kohms, and even higher when we have
a source resistance that is unbypassed such as the 30 ohms we have here.
The j-fet is like a pentode tube with a very high gm. For all tubes µ = gm x Ra.
And for all j-fets, µ = gm x Rd, so if Rd = 50k, then µ = 0.04 x 50k = 2,000, similar to many pentodes.
Usually engineers don't consider the j-fet amplification factor µ for j-fets since they know it is high, and
that Rd is also high, and variable with Id. J-fet gm is the all important parameter, but nevertherless,
if we know µ, then like a tube circuit the effective Rd with an unbypassed Rs = Rd + ( µ x Rs )
and in this case Rd' = 50k + ( 2,000 x 30 ) = 110k, very high in comparison to the cathode input
impedance of V1.
The j-fet is therefore a current source whose output voltage is determined by the load like a pentode,
and the gain is not affected by the Rd.
So in effect the V1 tube has  a very high value cathode resistance connected between its cathode and 0V.
Therefore the V1 tube is subject to a large amount of local current FB and it will work remarkably linearly.

THD in this cascode circuit is therefore primarily due from the transfer curve of the gm chartacter of the j-fet.
When the fet source R = 30 ohms, about 6dB of current FB is in action so the fet caused THD is reduced about 6dB from what it would be without the current FB.
Since the rated input signal is only 0.4mV at 1khz from my MC Denon 103R cartridge, the drain signal is only 4mV, and in fact THD is so very low it can be ignored.
If we wished to have the maximum possible output voltage from the j-fet to be 1.8Vrms into the V cathode
input impedance of 532 ohms, we would be dismayed to find THD was about 7%! ( without any current FB ).
But THD is about proportional to output voltage, so with 4mV, THD = 7% x 0.004/1.8 approx = 0.015%.
With the 6db of current FB this would reduce to around 0.008%.
But if the input signal was from MM at 4mV, drain signal would 40mV, and distortion couldn't be ignored
because it could be 0.08%.
So for higher level input signals the fet gain can be reduced by removing the gain link to increase the source
resistance from 30 ohms to about  250 ohms, and then the gain between gate and anode of the tube
will be 600 / ( 1 + [ 600 x 250 / 15,000 ] ) = 54.5, or about  15 db less than maximum gain, and because this is achieved by an  increase in the local current NFB, the THD of a larger input signal is controlled quite adequately.

The equivalant input noise resistance of a j-fet approximately = 0.7/ gm = 0.7 / 0.04 = 17.5 ohms in this case.
A triode's ENR = 2.5 / gm, so if the triode is a 12AX7, ENR = 2.5 / 0.0016 = 1,562 ohms.
The j-fet ENR = 0.0112 times the 12AX7's ENR.  The input noise varies in proportion to the square root of ENR, so that if the ENR difference is 0.0112, then the noise difference = sq.rt. of 0.0112 = 0.105.
Thus we could expect the j-fet to produce 1/10 of the noise of a good 12AX7, ie, the j-fet will be
20db quieter than the triode.
In practice, this is exactly what we do achieve by using a high transconductance j-fet, and in fact the noise produced by the j-fet  is far more benign than the triode since the triode noise includes flicker noise
which is low frequency and this sounds rumbly in a phono amp. The noise of the j-fet has little flicker
so the noise heard from the fet circuit is a very high pitched hiss, but it is at a level way below the triode noise.

This has been my experience with 2SK369. Although the type number indicates this tiny T092 package device is a mosfet, it really isn't, and just as well because mosfets have a lot more low frequency "popcorn" noise than j-fets, so choice of j-fet is quite critical. 2SK147 was made by Hitachi, but is now not made by anyone.
We rely on 2SK369 which is exactly the same as the 2SK147, but with a slightly lower power dissipation
of only 0.6 watts. In the circuit above the j-fet power dissipation at idle is only about 0.05 watts,
so the j-fet won't fail unless a short occurs between anode and cathode.
I bought a batch of 10 x 2SK369 for $1 each, so one can afford to replace them.
People say j-fets are fragile devices, and I would agree, but used in phone stages at such
low levels they seem to last well, but methinks the cap between gate and 0V, C3 needs to be there to shunt
any static discharges.  I have serviced Audio Research and Conrad Johnson preamps with
far more complex hybrid circuits than I am using, and these are difficult to diagnose
when a fault occurs, but my product is very easy to service.

I also found that amoung 4 randomly chosen samples of 2SK369, the gm matching was within 0.5%,
resulting in almost identical gains of the two stereo channels of the phono amp above and in another amp i built
for myself in 2004, mentioned in Preamps1 pages.
In one channel of the preamp in the above schematic there is a 50k linear trim pot for balance
adjustment. anyone could always use a dual gang linear pot for adjusting the gain of each channel to get a wider range of balance adjustment. Balance adjustment trimming is needed since the gains of V2 can be different even when one has bought a NOS pair of matched tubes.
I didn't think it necessary to have a formal balance control knob on the front panel.

The output impedance at the anode of V1 is extremely high, and approx = Ra + (µ x Rk)
where Rk is the impedance looking into the j-fet drain, perhaps 100k+, and µ = V1 amplification factor.
Since V1 µ = 33, the effective Ra of the 6DJ8 = 33 x 100k = 3.3 meg ohms.
But we have 23.5 kohms as the dc RL between anode and B+, so
at 5 Hz when the input impedance of the RIAA filter is very high, we can say that the
Rout from V1 = 23.5k in parallel with 3.3Mohms, or 23.33k.
So Rout from V1 anode = 23.33k with regard to calculations for the values in the passive RIAA filter.
If triodes such as 12AT7, 12AU7, 6EJ7, 6AU6 are used instead of 6DJ8, the effective Ro from V1 will not change substantially so no changes need to be made to the passive RIAA filter values once these have been established.

I found it extremely difficult to calculate the exact RIAA filter values but I did get approximate values
which I could start with. Then I used a reverse RIAA filter between my signal gene and the preamp,
and I adjusted the values to those I have in the above schematic by means of series/parallel arrangements of standard value resistances and capacitances.
Providing my reverse RIAA filter is correct, which I was told it was, so to will the preamp RIAA filter
after trimming R&C values. With persistance,  the reponse was brought within +/- 0.25dB of dead flat along the band with the reverse RIAA filter in place; the preamp output was a flat response using sine waves between
10Hz and 20kHz, and a square wave at 1 khz looked like a very nice square wave without ripples or sloped
horizontals.
My RIAA filter differs slightly to most other folks because it has an additional  R17 added in after the
3180uS and 318uS filter formed by R39, R18, and C7. The R17 + R19 + C8 act to give 75uS and about 3uS time constant filtering, and the R 17 builds the 75uS filter out from the R19 + C8 to help isolate the interactive effects. I find this arrangement was easier to get to within 0.25dB of the wanted flat response than when
ommitting R17 and using just a lower value cap from the top of R18 to 0V with a smaller value R in series for the 50kHz or 3.18uS time constant. Some people use 3 stage amp with RIAA 3180uS and 318uS attenuation done before stage 2, then they have the 75uS before stage 3, to get the most non interactive and accurate
RIAA eq, but here I have done it by simply adding some resistance between the 318uS and 75uS eq;
its simpler, and the V2 gain tube isn't subject to large HF signals; its input is a flat signal so there is less distortion.

The results of testing the accuracy of a phono amp will depend on the flat response of the signal generator sine waves and its low distortion. Its output resistance must be less than 100 ohms lest you get some errors at above 10khz. The Reverse RIAA filter has R3 = 3k3 to limit the amount of boost to HF above about 50kHz. Most practical cutting head amps have this extra HF time constant of about 3uS.

In my own 10 tube 2004 preamp I didn't use any B+ regulation. However in this  amp which I sold
I used an emitter follower style basic regulator at least for the B+ commonly supplied  to both channels.
This is in addition to the large value RC filter caps.
The emitter follower with BU208A plus MJE340 in a darlington pair was a simple way to achieve good ripple rejection and LF stability, and prevent cross channel talk.
The floating 12.6 Vdc applied to the heaters is shunt regulated by the two low voltage npn and pnp power transistors. The heater voltage is also biased and held steady with a cap to 0V.

V2, 12AT7, is the second gain tube with gain approx = 45. V2 in conjunction with the output follower tube V3
form what is a µ-follower configuration, with excellent linearity. If a sufficiently large 1khz input signal is applied
to the j-fet input to generate 10Vrms output from V3, THD = approx 0.2%, so that when output is only
0.5Vrms, THD falls to below 0.02% ( mainly all 2H ) and below the noise floor.

The use of parallel twin triodes for each triode element in the circuit gives excellent results.
I do not believe using paralleled triodes adversely affects the sound in power amps or preamps.
And I prefer to have only one signal function and one channel within one glass envelope.

There are those who would lock me in a dungeon and throw away the key for using a 12AT7 for ANYTHING!

They like to stamp on this tube with their jack boots. But I find that most precious prejudices against many
tubes are often just a lot of balderdash, and you can't  expect to get the same sound from all
brands and batches of 12AT7. If you don't try different brands of tubes, you won't know which are the best.
You may hear someone say they like Mullards or Siemans NOS, but you have to try them to see if they were right; 12AT7 isn't the most linear triode around, but we are only asking it to make less than a volt rms of output and I have used almost the most linear way of building a single ended voltage amp, so lack of linearity
is just not a problem, and anyone will be pleasantly surprised with the outcome.
While I rather like Siemans 6CG7, I am not so sure Sieman's 12AT7 have the same magic, and so perhaps a Mullard 12AT7, or 6201 might sound better since the Mullard sound, ( if there is a "Mullard sound" ) tends to
be polite, and kind to harsh recordings.  Because there is no global loop of NFB around the two gain blocks of this amp the changes of tube types that one makes will be able to be heard by folks with good gear in the
rest of their audio chain.

One interesting tube people could try for V2 is a 12AY7. It has similar gain and characteristics to a 6SL7, but is a 9 pinner and made for lower noise. However its anode current will be a lot lower than 12AT7 if just plugged
straight into this circuit instead of the 12AT7, and there will be too little current in the 6CG7 follower.
But for better operation with 12AY7, try R26 = 22k, R 25 = 1k approx so about 2mA of Ia flows at least,
then connect  about 100k between V3 cathode and 0V to increase the idle current in V3 for better
operation.
Of course if you can use a 12AY7, or 12AT7 for V2, then why try a 12AX7? It will also work OK
with the same slight mods as required by 12AY7. Maybe gain will be about 75 though, it could be too much.

I may be also thrown over a cliff soon when people see that I am no fan of having 10mA in each 1/2 triode section whenever I use a triode but honestly, I have never heard a big sound improvement when running
signal triodes in preamps at high temperatures due to high anode dissipation.  The use of high currents in signal triodes does place the tube operation into its region of more constant Ra and gm,
but when you analyse my schematic you will see that anode loads are high values, and the linearity
just does not suffer because of the operating points.

If ppl didn't want to use a 12AT7, they could use a 6DJ8 with a simple change of heater wiring
and perhaps a change to the cathode R25 and R26.

The V1 tube could be either 6DJ8, 6CG7 without any circuit changes. The V3 6CG7 at the output
could be a 6DJ8.

The tube choice does affect the sound quality probably more than any other changable factor such as coupling caps. I don't actually think there are better sounding caps than Wima polypropylene, 630v rated.
The topology used for each stage would also have a big effect on sound.  The V3 output followers
work with a lot of local NFB. Replacing brands of 6CG7 would not seem to be able to change the sound at all
because so much NFB is being used. All cathode followers have a lot of NFB, since all the output
is in series with the applied grid voltage, and gain is less than 1.0. The signal voltage between
grid and cathode is only approximately 1/18 of the output voltage at the cathode.
So one would expect the NFB to lessen sonic signatures caused by tubes with a different "voice", a word i hate to use, because in these sorts of amps the tubes don't have much voice, you only hear what was recorded.
There are no global loops of NFB in this phono amp, so the sonic signatures of the input j-fet,
V1 6DJ8, V2 12AT7 would add together give you what is heard.
The present j-fet is set up is one of several ways at least, and the V1 grounded grid triode is also
set up quite differently to more usual common cathode gain blocs. How can people form opinions about the sound of a particular type or brand of tube when their circuit topology can vary so much? So changing one of these devices in the chain is only changing the combination; it is impossible to attribute all the sound quality to any one device. Lady luck plays a part here.
But changing a tube still may make a slight sound change, maybe the sound will seem less bright,
warmer, whatever, its hard to predict an outcome without trying a few tubes.
I usually find that If I place well chosen tubes which test well to give low noise, correct gain, low microphony,
and low reverse grid current, then music really flows, and the sound is good if not extraordinary,
and changing tubes is a bit like opening a bottle of an alternative very fine wine; the taste may differ
but we can savour what comes from different wineries.

The new owner of this amp did confirm I'd proved my original point, that a few tubes and ONE j-fet
could sound better than a bunch of opamps from America.

The Rocket was supplied with Wima coupling capacitors initially. After some months with Wimas a trial of Auricaps was carried out with the line level preamp. One channel of the line stage had Wimas as originally supplied,
one with Auricaps. A mono source of signal from a CD player was used as a source and switched from one channel to the other. The mono output signal from either channel was then taken to both power amps and a mono signal
played from both speakers. Both myself or my client could not pick which channel had the Auricaps more than
50% of the time and I decided there was no difference in the sonic performance using either Wimas or Auricaps.

Please note that only the coupling caps in the signal path were changed to Auricaps. The bypassing
of power supply electrolytics and parts of the circuit prone to RF oscillations if left unbypassed
were bypassed with polyester capacitors. I believe such rail bypassing capacitors have absolutely zero negative effect
on the great sonic character of such tube amps. I remain quite doubtful that Auricaps can any beneficial effect
and will not be spending up to instal them in any of my own amps I use myself.

For power supply details see Nemo line preamp and power supply.

Back to index page.