=20
After more than 25 years of faithful service, it seemed that it might =
be time=20
to redo my phono system. After all, I like to think that I've picked up =
a few=20
tricks in the intervening years... The old system consisted of a VPI =
HW17-II, a=20
Linn Ittok LVII tonearm, and a Troika cartridge. The Troika was a =
substitute for=20
my previous cartridge, a Technics MM that I sorely miss. Given that the =
Troika=20
cannot be retipped for anything less than a California mortgage payment, =
new MCs=20
are priced like the fashion accessories that they are, and that the deck =
end of=20
the phono system still worked reasonably, my attention focused toward =
upgrading=20
the phono preamp and accommodating it to whatever cartridge I could dig =
up. That=20
ended up being a vintage Technics MC, which was a nice find- I'm a big =
fan of=20
Technics' high-end cartridges from the late '70s to mid-'80s. Second =
priority=20
went to bringing the VPI up to a higher standard, but that's a story for =
a=20
different time.
To celebrate the acquisition of the Technics, I =
decided=20
to just chuck the old phono preamp and build a new one from scratch. The =
old=20
one, a hybrid tube-FET cascode design, suffered from less-than-optimal =
noise and=20
distortion, insane power supply sensitivity, and OK-not-great RIAA =
conformance,=20
so the new one is designed to be better in those respects. And it turns =
out to=20
be simpler than my old unit, which is a bonus.
This article will =
describe=20
the preamp design and construction. An intermediate builder can =
duplicate this=20
circuit in a point-to-point or perfboard build, but if you're an =
advanced=20
builder, you've probably got your own way of doing things. Even if you =
don=92t=20
build this circuit, there are elements of the design process and result =
that you=20
might find useful in considering your own designs. Beginners will =
probably want=20
to use a PCB just to make sure that all is stable and quiet.
And, =
as=20
usual, I will make some very unpopular design choices. Let=92s go rile =
up the=20
villagers!
Background
May I let you in on a little secret? All = electronics are=20 not created equal. Some parts are more critical and difficult than = others. In=20 any analog system (and music is originally and ultimately analog), = there's some=20 sort of input transducer, some sort of output transducer, and a bunch of = wires=20 and circuit elements in between. The difficult bits, and the most = critical, are=20 where the output of the electronics meets a transducer (speaker or = headphones)=20 and where a transducer meets the input of the electronics (mike amps, = phono=20 amps, tape amps). And (confession) most of the in-between is pretty = trivial to=20 do well; levels and bandwidths are generally well-defined, and there's = not much=20 =93real-world=94 messiness.
The output part of the electronics = chain has=20 been well-handled by many fine power amp designs. In the tube world, the = phono=20 end has been much less successful. So-called classic preamps have = terrible=20 distortion and noise performance, poor drive capability, lousy headroom, = and=20 unmentionable overload recovery. Too many of the more recent designs are = just=20 lipstick on a pig; someone takes a circuit out of a tube manual, = designed for=20 the cheapest 1950s department store record consoles, slaps in a few = designer=20 caps, and proclaims it a wonder. Or a straightforward circuit is badly = crippled=20 by stuffing it full of fashion statements without regard to basic = engineering.=20 There are a few good ones out there, but not many.
I think we can = do=20 better. If you're going to do something as irrational as build a piece = of tube=20 gear, it's worth engineering it properly. And there are actually good, = sound,=20 rational reasons to use tubes for this application in the first place: = headroom,=20 overload recovery, and linearity.
Requirements
Accurate shooting requires a target. We will = start by=20 defining requirements:
1. Low noise. The cartridge chosen = has=20 chillingly low output (0.2mV/5cm/s) making this a nontrivial point. It's = unreasonable to expect 100dB signal-to-noise at these levels, but it = would be=20 nice to not add much to the cartridge=92s intrinsic noise.
2. = Tight RIAA=20 conformance. I don't think that 0.01dB absolute accuracy is = necessary, but=20 0.2dB is a reasonable target. Channel-to-channel should be at least = twice as=20 close (if not better), and this conformance ought to be robust toward = change in=20 tube characteristics with age.
3. Low distortion. This is = where=20 so-called classic phono amps really do a face-plant. One of the most = popular=20 "classics" drove a heavy capacitive load off the anode of a cathode = degenerated=20 12AX7, a very high source impedance. And because of the resultant = marginal gain=20 and the economic constraints of a two tube design, the designer threw in = a bit=20 of positive feedback. The mediocre performance, then, comes as no = surprise.=20
4. Freedom from overload and blocking. This is another = point=20 where "classic" circuits fall flat. Let the cartridge mistrack or rattle = slightly, whack the rock and lever that we laughingly call a "stylus and = cantilever," and that 50kHz ding-dong resonance kicks the input stage in = the=20 groin- remember, magnetic cartridges are velocity sensitive, and though = the=20 mistrack is only momentary, that tip resonance is a lot of velocity. A = minor=20 disturbance (a swift kick) can turn into hundreds of milliseconds of = overbias=20 (rolling on the ground, gasping for breath).
5. Capability of = driving=20 a 10k load and reasonable lengths of interconnect cables. The line = stage=20 with which this phono preamp is paired has a floating 10k input. No = reason that=20 a circuit should wuss out at this load just because it=92s tubes. =
6.=20 All-tube amplification. There are some decided engineering = advantages to=20 using tubes as the voltage amplification devices. At the top of the list = are=20 high linearity and overload immunity. My old unit was a hybrid using = FETs in=20 cascode with tubes (not unlike Allen Wright's excellent designs, though = not=20 claiming the same performance), and while it was good enough to live = with for a=20 quarter century, I still feel that its performance can be surpassed with = an=20 all-tube amplification lineup. Of course, I will still not hesitate to = use=20 semiconductors where they do best, i.e., for providing power, increasing = power=20 supply immunity, and controlling operating points- for constant voltages = and=20 constant currents, silicon is the way to go.
Design Requirement = 1 is the=20 toughest and most subtle one, and will drive some of the most painful = design=20 decisions.
A Short Digression on Noise
If you know your en = from your=20 in, you can pass over this section lightly.
There are = basically=20 two types of noise that electronic circuits can contribute, = deterministic and=20 random. Deterministic noise would include things like hums, buzzes, and = rattles.=20 These need to be fixed at the root-cause level, but they are fixable. = Random=20 noise is just that- random noise. Often white, sometimes pink, this is = the=20 =93shhhhhhhh=94 background that pollutes many phono systems. = Unfortunately, Nature=20 sticks us with an irreducible minimum of noise from any device or = source; the=20 amount of noise voltage that a device contributes is always greater than = or=20 equal to
(1) Vn =3D √(4KTRΔf)
where = R is the real part=20 (resistive) of the source impedance, Δf is the bandwidth over which = the voltage=20 measurement is taken, K is Boltzmann=92s constant, and T is the = temperature or=20 equivalent temperature. The derivation of this equation from first = principles is=20 shown in the Feynman Lectures on Physics, and is one of the most = beautiful=20 results in all of science. Admire it. Worship it. This equation will be = with us=20 until the end of the Universe because it is a fundamental consequence of = the=20 structure of matter.
Likewise, any random noise can be = represented as an=20 equivalent resistance, that is, for a random noise voltage of V measured = over a=20 range Δf, there is an equivalent ideal resistor R at a temperature = T that is=20 given by
(2) Req =3D = V2/(4KTΔf)
The takeaway:=20 noise and resistance are correlated and can be inter-converted as = equivalents. A=20 random noise source can be expressed as an equivalent resistance and = vice=20 versa.
Don=92t get too fat and happy. This theoretical value is = the minimum=20 noise that a real-world component can give. Real components add both = random and=20 systematic noise to the ideal resistance. This noise is often called = (logically=20 enough) =93excess noise.=94 For example, if you measure the noise from = carbon, metal=20 film, and wirewound resistors of the same value, you=92ll find that the = wirewound=20 resistor will have noise very close to ideal, the metal film will be = slightly=20 worse, and the carbon resistor to be a bit worse yet (sometimes a LOT = worse).=20 And to add to the worries, that excess noise is dependent on current=85 = Well, we=20 have our eyes open.
Because the source noise is random and the = various=20 sources of noise are uncorrelated, generators in series add as noise = power=20 rather than noise voltage, i.e., square root of the sum of the squared = noise=20 voltages. So for example, suppose one has three uncorrelated noise = sources in=20 series of 1mV, 0.5mV, and 0.1mV. The total equivalent noise voltage is = then=20 √(12 + 0.52 + 0.12) =3D 1.12mV. = It=92s notable and=20 convenient that the noise power addition does make the largest voltage = noise=20 source overwhelmingly dominant, hence one doesn=91t need to do terribly = precise=20 calculations to get a good estimate of noise in complex circuits-=20 simplifications will help.
A nice little calculational shortcut: = if we=20 plug room temperature and a 20k bandwidth into the noise equations, we = can=20 express the noise voltage-resistance relation as
(3) = Vn =3D 1.8=20 x 10-8 √R
Or
(4) R =3D = Vn2 (3.1x=20 1017)
And because of the RIAA equalization, any=20 pre-equalization noise is reduced by the downward slope of the transfer=20 function. The math is handled very nicely in a National Semiconductor=20 application manual (Audio Radio 1980, Appendix), but the bottom line is = that for=20 a white noise source (characteristic of resistors), the effective = bandwidth is=20 reduced from 20kHz to 120Hz. So since V is proportional to the square = root of=20 bandwidth,
(5) VRIAA =3D Vn √(120/20k) =3D 1.4 = x 10-9=20 √R
This approximation is surprisingly accurate.
So = a noise source=20 can be expressed as an equivalent resistance R or as a generator voltage = Vn. There=92s one more little twist- sometimes, noise is = expressed as=20 =93noise voltage density,=94 usually symbolized by en. = Usually given in=20 units of voltage per square root Hertz, it=92s a way of comparing noise = sources=20 without putting bandwidth into the equation. For example, an ideal 1k = resistor=20 will (by equation 3) have a noise voltage from 20Hz-20kHz of 5.7nV. = Another way=20 to characterize the noise is to use equation 2, and = rearrange:
(6)=20 en =3D V/(√Δf) =3D √(4KTR)
Using this = formulation, the 1k resistor=20 is said to have a noise density of 4.1nV/√Hz. One can multiply = this value by the=20 bandwidth to get a noise voltage. Using voltage noise density is useful = for=20 situations where the noise source is not a simple resistance and where = the noise=20 voltage density changes with frequency.
As an interesting = exercise, we=20 will compute the signal to noise of the cartridge itself, just to see = how quiet=20 the preamp needs to be in order that it does not add any significant = noise. The=20 cartridge resistance is 15R. Plugging that value into equation 1, we = find that=20 the cartridge has 72nV of thermal noise at room temperature over the = usual=20 20Hz-20kHz audio bandwidth. The cartridge=92s nominal output is 0.2mV. = Doing a=20 quick decibel check, signal to noise works out to -69dB. Remember that = number,=20 it will come up again.
*
Overall Topology
There are several approaches to low noise = tube RIAA=20 stages. One is the classic feedback circuit popularized by Dynaco, = Marantz, and=20 Audio Research. These circuits are=85 not great. The limited open loop = gain means=20 that RIAA conformance changes with tube aging or replacement, the first = stage=20 will require a noisy cathode resistor, and the output stage will drive a = heavily=20 capacitive load at high frequencies. Distortion rises at the low end, = where the=20 feedback isn=92t very effective, and at the high end, where the = capacitive load=20 reduces the available open loop gain and in some cases causes slewing=20 distortion. What was it that Morgan Jones said about classic phono=20 stages=85?
Since we=92re working with tubes, which are = intrinsically more=20 linear devices than transistors and can swing considerably more voltage, = there=92s=20 no reason not to use a passive equalization scheme. One quite simple=20 implementation is the topology shown in Figure 1. In this topology, a = so-called=20 =93all-in-one=94 RIAA equalization network is sandwiched between two = gain stages. A=20 buffer isolates the second gain stage from the load (interconnects and = whatever=20 you=92re using for a line preamplifier). The all-in-one RIAA network = achieves the=20 3180us, 318us, and 75us time constants in the RIAA standard; working = equations=20 for this circuit are given in a useful paper by Lipshitz, or you can use = a=20 handy-dandy on-line calculator which may be found at
A common variation of this = topology is=20 splitting the 75us time constant off from the other two, then = sandwiching it=20 between the second gain stage and the buffer. In many cases, this can = help with=20 noise and headroom, but I have a sneaking suspicion that it=92s done = that way just=20 to make the math easy. No matter, if the first stage has LOTS of gain = and=20 reasonably low source impedance, the all-in-one network will carry no = noise=20 disadvantage and will certainly improve the headroom of the second gain=20 stage.
Figure=20 1. Passive EQ RIAA stage topology
Let=92s estimate the gain = we=92ll need=20 overall, taking into account the RIAA equalization. At 1kHz, RIAA is = down 20dB=20 from 50Hz. So if we want (nominally) 0.5-1V out from the phono preamp at = 0.2mV=20 and 1kHz, we=92ll need about 85-90 dB of gain (taking into account the = loss from=20 RIAA). No wonder so many circuits fall flat! Now, because the 0.2mV spec = is at=20 5cm/s groove velocity and many records have modulation higher than that, = we can=20 relax the gain a few dB and make it up in the line stage, if = necessary.
First Stage
OK, I'll begin by turning off half my readers- = the=20 design will use an input transformer. This "impure" approach carries too = many=20 advantages to ignore, including nearly noise-free gain, galvanic = isolation, and=20 the opportunity to run the phono cartridge in a balanced mode without = doubling=20 the tube count. Running balanced inputs is an amazingly powerful = technique to=20 reject common-mode noise and is common studio practice for low-level = mike=20 signals. This does require a nonstandard cable, but some twisted pairs = and=20 shields never hurt anyone.
The question begged is, =93Can I run = an MC in=20 directly?=94 Mmmm, perhaps, but at a noise penalty. Let=92s say that you = choose a=20 particularly quiet specimen of tube; the very best have input-referred=20 equivalent noise resistances in the 60-100R range. Cherry pick to get a = 60R.=20 Compared to the cartridge thermal resistance (15R), that represents a = 7dB=20 degradation of the already-marginal signal-to-noise ratio. =
Massive=20 paralleling of input tubes might prove efficacious, at a stunning = penalty. Four=20 tubes in parallel will drop the equivalent resistance to 15R, reducing = the=20 degradation to 3dB, which is still not quite there. Transformer starting = to look=20 better?
Things get worse: to get that low noise out of a tube, = the=20 transconductance will be quite high, and high transconductance means = oscillation=20 is but a moment away. The solution is generally grid-stopper resistors, = but they=20 have to be 100R-1k before doing any good, providing yet another noise = source,=20 one big enough to swamp the tube=92s noise. Maybe you can wave the magic = ferrite=20 beads on the grid leads and keep things stable (I can=92t). But that = just takes us=20 back to the tube noise problem. Nope, if we want tubes, we want MC, and = we want=20 quiet, we need a transformer.
Once we have crossed that Rubicon, = we find=20 that the transformer brings some intrinsic advantages of its own. One = advantage=20 that do-it-yourselfers have over appliance operators is the ability to = use=20 non-standard interfaces if that will get the job done. We can design and = construct for performance and not worry about some strange item that = might get=20 plugged into either end. This can pay dividends if we consider the phono = cartridge, arm, cable, and preamp as a system instead of worrying that = each part=20 be universal. Out in the Real World, where billions are made or lost = depending=20 on the engineering, balanced circuits are de rigueur for transporting = low level=20 signals from Point A to Point B. So where do audiophiles use them? Why, = between=20 preamp and power amp, where signals are large and hum is easy to = prevent. And=20 high enders love doing it inside preamps, where there=92s shielding, = short wire=20 runs, and controlled grounding. The weak link, the cartridge to preamp = transfer=20 of microvolts of precious signal is of necessity outside and exposed to = the=20 cruel world. Yet it is almost invariably done single ended, making a = mockery of=20 all further effort. It becomes doubly incredible when one considers that = a=20 cartridge (the odd Decca excepted) is not inherently single-ended; it = can have=20 both ends float just as well as it can have a leg staked to the ground.=20
The phono cartridge unquestionably needs to be run balanced. The = signals=20 are tiny, the opportunities for hum and noise pickup are huge. As a = practical=20 matter, that involves changing two hunks of wire and designing a phono = stage=20 with a balanced input. The first hunk of wire is the tonearm wiring. In = a many=20 cases, it might be thin, twisted wire, which would be ideal for balanced = connection. If not, the intrepid constructor will then need to replace = the arm=20 wiring. Looking at what that would involve for my rare Linn arm, I = decided that=20 I was not intrepid, and refused to take the arm apart to check the = wiring. But=20 please do as I say, not as I do. As it happens, the short length of = uncontrolled=20 wiring in the arm didn=92t cause me any undue noise problems, but every = listening=20 room is different.
The second hunk of wire is the = tonearm-to-preamp=20 cable. This is a critical hunk, since signals are small and noise lurks = around=20 every corner. One excellent suggestion from Morgan Jones is to sheath = silver=20 wire in Teflon sleeve, twist it, then slip a shield braid over the whole = shebang. One of these days, I=92ll do that, at least if I win the = lottery and get=20 over my fear of triboelectricity. In the meantime, I picked up a Good = Trick, and=20 found that some shielded CAT5 cable worked well. That stuff is perfect, = several=20 sets of twisted pairs, good quality wire and insulation. The pairs can = be=20 paralleled to reduce the cable resistance. And I can spend the = difference=20 between that and the silver/Teflon approach on beer and lottery=20 tickets.
Once that balanced signal is delivered to the preamp=92s = input,=20 the last remaining link is the Common Mode Rejection Ratio (CMRR) of the = input.=20 In order to do any good, a balanced signal needs a balanced input on the = preamp.=20 That input ought to have as high a common mode rejection as possible. = The=20 classic single-ended tube input will contribute nothing. A balanced,=20 differential input will be better, but doubles the input stage noise. = The=20 transformer still seems to be our best choice.
My old preamp used = vintage=20 Peerless/Altec 4722 mike transformers, which gave a very healthy 1:32 = stepup.=20 Unfortunately, this came with a bandwidth penalty because of the = interaction of=20 the secondary with its own capacitance and the input capacitance of the = first=20 stage. Thus my previously-mentioned Hobson's choice of a FET cascode, = since=20 first stage input capacitance for any high mu triode is too high for = this=20 transformer to handle without some impressive ringing of its own, right = around=20 where the cartridge is misbehaving the same way. That's not = good.
Better=20 transformers are available, and one excellent choice is the 1:10 Sowter = 8055X,=20 which was designed to have excellent input balance. The big disadvantage = for the=20 American builder is the price - after bending over for the unfavorable=20 dollar/pound exchange rate, then paying shipping and duty, the Jensen=20 equivalents start looking much better. Serendipitously, a friend of mine = in the=20 process of moving ran across a spare pair of the Sowters, which he sold = to me=20 for a bargain price.
The Sowters seem much happier with = capacitive loads=20 than the Peerless units, and on the test bench, I found that from a = source=20 impedance of 15R driving the primary, a 6k8 secondary load resistor = paralleled=20 with 200pF across the secondary gave me rather beautiful square waves, = free of=20 ringing, and with a 4us rise time. That gives me quite a bit of = flexibility as=20 regards input stage capacitance- the FET can be dispensed with, and I = don't need=20 to suffer the disadvantages of cascodes (like their essentially zero = power=20 supply rejection). And the common-mode rejection was measured to be in = excess of=20 100dB at 60Hz. All right! To make things even nicer, the secondary = resistance=20 (which determines the noise contribution from the transformer) is quite = a bit=20 lower than the competition- about 100R compared to nearly 1k in the = Jensen 1:10=20 step-ups. This results in about a dB of lower noise, but every dB = counts,=20 doesn=92t it? The nifty little attenuator I built for transformer = testing is shown=20 in Figure 2. It=92s driven from the balanced output of my M-Audio = Audiophile 192=20 soundcard, knocks down the signal level 74dB to avoid stunning the = transformer,=20 and has close to 15R source impedance like the cartridge. If you=92re = using a=20 different cartridge (likely) or a different transformer, you can = transform the=20 values I used accordingly.
Figure=20 2. Attenuator for transformer tests
The 6k8 load does knock = down the=20 cartridge's output a bit, about -1.5dB to 0.15mV, but the signal to = noise=20 degrades less than that since the effective Thevenin resistance of the = input=20 system also drops from 1k6 to 1k3. But the loading still costs us a = decibel of=20 S/N.
With the transformer present and accounted for, the next = most=20 critical decision is the nature of the active part of the first stage. = If we=20 don't get it right here, we won't be able to recover later. And a good = choice at=20 this point will ease the overall design. The basic requirements here are = low=20 noise, low noise, and low noise. Secondarily, we would wish the highest = gain=20 possible- the input transformer has been kind enough to give us 20dB and = we=20 wouldn't want to let it down. We also recognize that the necessary RIAA=20 equalization will knock down the level at 1kHz by about 20dB, so we're = starting=20 out on the wrong end of the lever. Every bit of gain will = help.
The=20 cartridge has a source resistance of 15R. This is transformed up by a = factor of=20 100 to 1k5 by the 1:10 ratio of the input transformer. The input tube = then=20 should have an equivalent noise resistance that=92s low compared to 1k5 = so as not=20 to add significant noise. It should also have a mu as high as possible = to get=20 the still-tiny signal out of the muck. That turns out to be 70-100 for = practical=20 triodes. One can get higher gain from a cascode, but an all-tube cascade = that=92s=20 linear and stable is not a trivial exercise; worse yet, cascodes are = superb at=20 passing along every last bit of power supply noise, making the design = exercise=20 in ultra-low noise supplies a complex one. I think we can do as well = with a=20 classic grounded cathode triode input stage- not tricky, not glamorous, = but it=20 works.
A popular tube for this position is the 12AX7/ECC83. That = can be a=20 decent choice but requires quite a bit of design thought because of the = high=20 plate resistance (60-80k). If the passive RIAA build-out resistor = isn=92t large=20 compared to the tube=92s source resistance, the RIAA conformance will be = dependent=20 on the tube- not a good thing as tubes warm up, age, and change. A large = build-out resistor means noise. Also not good. The 12AX7 often gets a = bad rap=20 for linearity, but the reality is that, with a very high plate load = resistance=20 (or better yet, a CCS), it has stunningly good linearity. But it does = need that=20 high plate load. Even with CCS loading, at high frequencies, the = tube=92s=20 effective plate load becomes the RIAA build-out resistor, one more = reason that=20 the resistor has to be large and noisy.
One other tube = traditionally used=20 in the first hole is the 417A/5842. This tube has a very low plate = resistance=20 (1k6) and high transconductance (25mA/V), but the mu is marginal (43). = If only=20 we could combine the low plate resistance of the 5842 with the high mu = of the=20 12AX7=85
And of course, we can. The D3a is a European pentode = that has=20 become much better known in the past several years. It=92s easily = available, not=20 terribly expensive ($10 is average), and can be connected in triode mode = to give=20 us a tube that=92s ideal for this application. Mu in triode connection = is 73 (not=20 quite as good as a 12AX7, but 4.6dB better than a 5842), plate = resistance is=20 slightly over 2k, and the transconductance is an impressive 35mA/V. = Equivalent=20 noise resistance is 65-100R, certainly well below the 1k5 transformed = source=20 noise. It is slightly tricky to use- the high transconductance means = that layout=20 is critical and stopper resistors must be used to prevent oscillation. = And=20 there=92s a lot of unit-to-unit variation, so it=92s worth getting = extras and doing=20 some selection.
For the grid-stopper, I found that 100R kept = things=20 calm. And the noise contribution (compared to the 1k5 resistance of the=20 transformed cartridge) is negligible.
Let=92s choose the = operating point.=20 Looking at the D3a datasheet, we see that both plate resistance and mu = vary=20 rather steeply with current at low currents. By 20mA, they are beginning = to=20 level off, so let=92s use that as a starting point. That leaves plate = voltage and=20 grid-cathode bias as the remaining interdependent variables. The bias = voltage=20 will determine the overload characteristics of the stage, so we want = that to be=20 high. But that also forces the plate voltage to be high and threatens = excessive=20 dissipation. For reasons that will soon be clear, 1.2V grid-to-cathode = will work=20 well. This results in about 140V on the plate- at 20mA, that=92s a bit = under 3W=20 dissipation, so the tube is run well under its 4.5W limit and should be=20 reliable.
Overload at the grid will probably start about 1V peak, or = 0.7VRMS. Will that be sufficient? The cartridge has a 0.2mVRMS nominal = output at=20 5cm/s. That ends up 1.5mV at the grid. Mistracking, dirt, and other = vicissitudes=20 of the LP Life will whack the stylus like little hammers, which is a = lovely way=20 to excite the high Q ultrasonic resonances to which all MCs succumb. The = cartridge is a velocity transducer, so the high frequencies generate=20 proportionately higher voltages. It=92s not unreasonable to want at = least 20dB of=20 headroom above that. With our chosen operating points, we have 53dB = before any=20 input problems. I think that will do.
Now the question of bias = method.=20 Cathode bias using a resistor is a common method. But it carries the = penalty of=20 reduced gain and increased effective plate resistance. Worse, the = cathode=20 resistor contributes its own noise- the cathode is, after all, an input=20 terminal. Bypassing can help, but requires a large cap, almost certainly = an=20 electrolytic. That can=92t be good. And finally, although it=92s = unlikely this stage=20 will overload, bypassed cathode resistors turn brief overloads into = severe and=20 clearly audible =93choking=94 of a stage by extending the recovery time. =
Another solution is battery bias. This is quite a good one, but = I just=20 don=92t trust the stability and reliability of batteries, especially = wrapped up in=20 hot boxes full of glowing tubes.
My favorite solution, as anyone = who has=20 seen my earlier projects will know, is LED bias. The origins of this = clever idea=20 are obscure (I first saw it in the late =9170s, proposed by Ike = Eisenson), but it=20 works like a charm. Forward biased LEDs have low dynamic impedance, low = noise,=20 high bandwidth, and provide essentially instantaneous recovery from = overloads.=20 They=92re also nice troubleshooting devices- an LED that=92s lit means = the tube is=20 conducting. The dynamic resistance is a function of current, so the = magnitude of=20 that and its effect needs to be considered in the design. The noise of = the IR=20 LED was too low for me to measure, probably somewhere around the Johnson = noise=20 of its dynamic impedance; even if the equivalent noise resistance were = ten times=20 the dynamic impedance, it would be negligible.
Figure=20 3. First stage topology
OK, we have our 1.2V bias, we=92re = running 20mA=20 through it, we know from the datasheet that the plate voltage is going = to be=20 140V, the last bit is the plate load. We can determine a good one by = putting a=20 dot on the D3a plate curves and pivoting a ruler around to find a good = load=20 line. But the most linear of all is a perfectly horizontal line, that = is, a=20 constant current. Constant current also maximizes the gain to near mu.=20
Constant current sources as plate loads bring some other = advantages to=20 the party. Constant current means that the variation in LED dynamic = impedance=20 with signal can be safely ignored. The plate voltage is automatically = adjusted=20 tube to tube and as tubes age to maintain the correct operating point. = And power=20 supply rejection increases dramatically- the CCS acts like an enormous = (100M or=20 more) series resistance, and that resistance forms a voltage divider = with the=20 tube=92s plate resistance. So any noise from the CCS or the power supply = rail is=20 knocked down another 80-90dB or more.
The best bang-for-buck CCS = also=20 happens to have exceptionally high performance. The DN2540 depletion = mode MOSFET=20 is perfectly suited to make a simple cascode CCS with output impedances = north of=20 100M and exceptionally low noise (14pA/√Hz, an insanely minuscule = amount when=20 multiplied and integrated with the D3a=92s low plate resistance). So our = basic=20 gain block will look like the circuit in Figure 3.
We will check = the=20 input capacitance. The gain of the stage will be about 73. The grid to = cathode=20 capacitance is 7pF, the grid to plate capacitance is 2.7. Plugging gain = and the=20 latter capacitance into the Miller equation, then adding in the former=20 capacitance and another 5pF for strays, we end up with about 210pF. Say, = wasn=92t=20 that what we wanted to load the input transformer=92s secondary? Hmmmm, = quite a=20 coincidence=85
This exegesis on the first stage may seem overly = long, but=20 it=92s the single most critical part of the design- anything wrong here = will be=20 faithfully passed down the chain and can=92t be fixed.
Let=92s = see where we=20 are: the input-referred equivalent resistance of the tube is 65R. Source = resistance of the transformed cartridge plus transformer secondary = resistance=20 plus grid stopper is 1k5 + 100R + 100R =3D 65R =3D 1R765. From equation = 1, this is=20 equivalent to 0.76uV. Our 0.2mV signal has been transformed to a 1.5mV = signal by=20 the transformer, so our signal to noise is -68 dB. Remembering that the=20 cartridge=92s thermal noise limits the maximum obtainable signal to = noise to=20 -69dB, we=92ve gotten through the first stage relatively unscathed! Our = signal is=20 now approximately 0.15mV * 10 * 73 =3D 113mV, which is well out of the = muck and=20 something we can deal with. At that swing, the distortion from the D3a = drops to=20 below any reasonable measurement limit. In the spirit of democracy, = let=92s move=20 on and equalize!
The EQ Network
This is the easiest part. Once we make one = basic=20 decision, it=92s all rote calculation. The D3a stage has an output = impedance of=20 about 2k2 or so. Referring back to Figure 1, R1 comprises the source = impedance=20 of the D3a. Adding a series resistance will make the loading on the tube = kinder,=20 limit the drift of EQ accuracy with tube aging, and provide a convenient = way to=20 trim the network for accuracy. If we choose the value of R1 to be about = 10 times=20 the plate resistance, the gain will hardly budge and neither will the=20 distortion. A 10% drift in the tube=92s plate resistance will only cause = a 1%=20 change in the effective value of R1. The only penalty is slightly more = noise. We=20 can take comfort in the observation that many well-regarded designs use = much=20 bigger (and hence noisier) resistors in this position, but we=92ll = quantify that=20 momentarily. By jiggering things around a bit, we can try to get as many = standard values as possible. With the total R1 (resistor plus tube) of = 21k7, C1=20 works out to be 0u1, R2 to be 3k15, and C2 to be 33n5. Given the = tube=92s output=20 resistance, R1 will be somewhere north of 18k- I used a 20k, then used = large=20 resistors (>200k) in parallel to trim it. Likewise, R2 can be 3k3, = with a=20 large resistor (100k) in parallel to trim it into place. C2 came very = close to=20 33n, a standard value, and could also be trimmed with a few hundred = pF.
Figure=20 4. Thermal noise from RIAA network
Now, what will be the noise=20 contribution of this network? In this case, because of the shunting = effect of=20 the various capacitors, the voltage noise density will have a strong = frequency=20 dependence; it won=92t be just be R1, it will be somewhat smaller = because of the=20 parallel reactances of C1 and C2. There=92s at least two ways to handle = the=20 complication of the shunt caps. If you=92re a He Man, you=92ll follow = the procedure=20 that National Semiconductor outlined in the early versions of their = Audio and=20 Radio Applications handbook. This involves dividing up the audio = spectrum into=20 frequency bands, computing noise from each band, then power summing the=20 contributions.
The wuss way (but far more accurate) is to send a=20 schematic to someone who can actually use SPICE (I am totally inept) and = have=20 the computer run a much more accurate simulation. Being a wuss, I pawned = the=20 task off onto a Dutch elf and ended up getting the graph of Figure 4, = which=20 shows en as a function of frequency.
To determine how = much=20 effect this noise source has on our circuit=92s signal to noise, we = unfortunately=20 have to do some He Man math (though in truth the computer could have = done this=20 for us). First, note that because this is a log frequency plot, the rise = at low=20 frequencies has very little effect- Δf is pretty small. No matter, = let=92s do our=20 sums. Divide the plot into three segments, 10-100 Hz, 100-1000 Hz, and=20 1000-20,000 Hz. For the first band, en runs between 10 and 25 = nV/√Hz.=20 Let=92s approximate it by 15 nV/√Hz. Δf is 100 Hz, so the = voltage contribution=20 from that band is 0.15uV. The next band has en at about 6 = nV/√Hz. Δf=20 is 900 Hz, so the noise voltage contribution is 0.18uV. And finally, the = last=20 band averages out to something like 2 nV/√Hz with a Δf of = 19,000 Hz, for a noise=20 voltage contribution of 0.28 uV. Total noise from the network is then = the power=20 summation of the three sources, or 0.37uV. We have 113mV of signal which = is=20 knocked down 20dB at 1kHz by the RIAA EQ, or 11.3mV. The noise voltage = from the=20 RIAA network is then seen to be better than 90dB down from the signal. I = think=20 we need not worry about the RIAA network=92s contribution to the noise!=20
There=92s one little twiddle to this that we=92ll implement in = the final=20 circuit; tune in later for The Case of the Missing Zero.
The Second Gain Stage
The hard part is done- the signal is = of decent=20 size, the equalization has been implemented, and now all we need to do = is get=20 the signal just enough bigger that we can hand off duties to the line = amp. If we=20 wish the nominal 0.15mV signal at the input to give us something like = 0.5V on=20 the output (leaving room for =93hot=94 cut records to go significantly = higher), then=20 the 11.3mV signal at 1kHz will require the next stage to have a gain of = about=20 35-40. This stage will be handling the biggest input signals and have to = swing=20 the most volts at its output. Simply because it will give a gain of 35 = and I=20 have had a lot of positive experience with it, the next tube in the = chain will=20 be a 6DJ8/ECC88 or one of its variants. These tubes are relatively = inexpensive=20 ($10), quite linear, and have an equivalent noise resistance of = 200R-250R, well=20 below the input noise.
Being the lazy sort, I will use the same = topology=20 as the first stage, though the bias voltage will be larger in order to = give a=20 bit more headroom. From previous work, I found that with 10mA of current = and=20 1.7V of bias, the ECC88 was at a very nice linear point, with about 90V = on the=20 plate. So the LED in the cathode becomes a red one (1.7Vf) and the plate = CCS is=20 adjusted for 10mA. And that=92s pretty much it.
A few minor = details,=20 though. First, because the transconductance is fairly high and the=20 interelectrode capacitances are fairly low, the ECC88 will oscillate if = you give=20 it half a chance. So, don=92t go without protection- use a grid stopper, = preferably as tight as possible to the tube=92s grid pin. Second, = there=92s the=20 question of coupling the first stage, the RIAA network, and the second = stage=20 together. One popular method is RC coupling right after the first gain = stage and=20 before the RIAA network. This has the advantage of keeping high DC = voltages off=20 of the RIAA components. The disadvantage is that the network is now = driven by a=20 source that changes impedance with frequency at the low end. This can be = overcome by using a HUGE coupling cap, but why bother? Let=92s put the = RC coupling=20 after the RIAA network and spend the (perhaps) extra dime using 400V = caps in the=20 RIAA.
In order that the RC coupling not load down the RIAA = network and=20 attenuate the signal, we want a nice, large grid leak resistor. 1M is a = safe=20 value and barely disturbs things. The usual 0u1 coupling cap completes = the=20 picture, and is large enough that the 1M resistor is shunted by the much = lower=20 impedance of the RIAA network at frequencies above 100Hz, so contributes = pretty=20 much diddly squat to the noise.
Finally, a key requirement of = this stage=20 is overload immunity- we don=92t want blocking. Fortunately, the RIAA = network=20 really hammers down the treble frequencies most likely to cause an = issue. And in=20 the midband, the overload margin is ridiculously high: 11.3mV versus = slightly=20 more than 1.6V to cause overload- that=92s 40dB of cushion. You won=92t = see anything=20 like 0.15mV from the cartridge at low frequencies, but even if you did, = that=20 translates to 120mV input to the second stage, well below the overload=20 point.
Well, that was easy, wasn=92t it?
The Output Buffer
There=92s a temptation to take signal = right off the=20 second gain stage plate. After all, the ECC88 has a plate resistance of = about=20 3k=85 Remembering Design Requirement 5 (ability to drive interconnect = cables and a=20 10k load) should give us pause. Putting a reactance in the plate load = and=20 swinging the load line vertical will significantly reduce the gain (from = 35 to=20 26) and increase the distortion. We don=92t want that, do we? Of course = not, so=20 it=92s probably a good thing to insulate that tube from the vicissitudes = of the=20 Real World by attaching a buffer.
I=92ve made no secret of my = affection for=20 a properly designed cathode follower. Though there is a quasi-religious=20 objection to this simple and wonderful circuit, the fact remains that = not only=20 does a cathode follower measure nearly perfectly, but no one (and that = means NO=20 ONE) has ever demonstrated that they could hear the effects of a = competently=20 designed cathode follower inserted into a signal chain. The objections = are=20 either from experience with badly designed followers (and there are many = of=20 those, sadly) or theoretical philosophy, which gives me heartburn. = I=92ve tried=20 fancier circuits (bootstrap, mu follower, White follower) and never = found one=20 that actually worked better for the requirements of an audio = preamp.
So=20 with simplicity our key, I=92ll refer you to my article on The Heretical = Preamp=20 for an overly detailed analysis of the Right Way to design a = follower.
Part=20 of the Right Way is using a high transconductance tube, since follower = source=20 impedance is inversely related to transconductance. Another part of the = Right=20 Way is using a current sink to set operating points and maximize the = follower=92s=20 load. Let=92s consider each of these in turn.
First, tube choice. = We=20 already used an ECC88 as the second voltage amplifier- we have a section = left=20 over, so why not use that as the follower? Transconductance is good, and = the=20 Heretical showed that with a CCS in the cathode, the distortion = performance of=20 the ECC88 is impeccable. How convenient!
Now, we have two good = choices on=20 how to set up the follower. One method is to direct couple from the = plate of the=20 second voltage amplifier to the grid of the cathode follower. This puts = the=20 cathode some 90V above ground, plenty of room for the CCS load to = operate. The=20 output is then capacitively coupled. An alternate method is to = capacitively=20 couple from the plate of the second voltage amplifier to the grid of the = cathode=20 follower, then return the CCS load to a negative voltage rail. This = allows the=20 output of the preamp to be direct-coupled and servoed, a la the = Heretical. The=20 advantage of the second method is that it replaces a large coupling = capacitor on=20 the output (on the order of 1 uF to keep the LF rolloff below 2 Hz) with = a much=20 smaller capacitor on the follower input (on the order of 0u01 for the = same=20 rolloff). The disadvantage is the need for a negative rail and = considerably more=20 complication in the circuitry.
I've opted for the first method = due to=20 simplicity, but would not argue with anyone who wanted to implement the = second=20 method instead. Either way, the stage will have a source impedance in = the low=20 hundreds of ohms, and a set of interconnects and 10k at the far end will = not=20 faze it a bit.
The blocks are arranged and outlined. Time to = look at the=20 whole package and put together a finished design.
Design Integration and Details
The schematic of the signal = section=20 is shown in Figure 5. The blocks will look completely familiar, but = there's a=20 few details that need explaining.
The implementation of the CCS = loads is=20 identical to that of the ImPasse preamp (I'm lazy and prefer to re-use = good=20 circuit blocks in all kinds of places). And like the ImPasse, each CCS = is held=20 at arms length from its tube by a resistor (R3 and R12). R5, R14, and = R18 set=20 the CCS currents, 20mA for the first stage, 10mA each for the next stage = and the=20 buffer. The values may have to be adjusted slightly, depending on the=20 particulars of your batch of MOSFETs; this is most easily done by = attaching the=20 + end of the CCS to a power supply (24V or more) and the =96 end to a = 100R dummy=20 load resistor connected to ground. The 20mA CCS should be adjusted to = get 2V=20 across the dummy load, and (naturally) the 10mA CCS adjusted to get 1V.=20 Exactitude isn't critical.
Figure=20 5: Signal Circuitry
Next question- what's R9 doing in = there? This=20 part is optional- some claim that there's a missing zero at 3.18us due = to the=20 rolloff of the Neumann cutter heads used to cut many records. Huge sonic = advantages are claimed by adding this Missing Zero. Others disagree. = Frankly, I=20 can't hear a difference with the resistor in circuit or shorted out = (though=20 admittedly my old ears would be lucky to hit 15kHz on a good day), but = am=20 leaving it in the schematic as an option for those who believe that they = CAN=20 hear it.
The neon bulb between cathode and grid of the follower = protects=20 the tube at turn-on by limiting the voltage between grid and cathode to = about=20 70-80V. This isn't quite as safe as the usual reverse-biased diode, but = it's=20 probably below the flashover point and doesn't have the nonlinear = capacitance of=20 a reverse-biased semiconductor diode. The neon will fire at turn-on, = then go out=20 as things start to warm up.
The output features a cathode stopper = which=20 helps ensure stability at the expense of slightly higher output=20 impedance.
Q2 dissipates the most power of any of the devices so = should=20 get its own little heatsink. You can do the same for Q4 and Q5, but it's = not=20 quite as critical.
Pretty straightforward!
We move to the=20 regulator circuits, shown in Figure 6. For the HV regulators, four are = needed,=20 two at 260V, two at 160V. This allows each channel to have separate = regulation.=20 Don't separate the feeds for the second gain stage and the cathode = follower -=20 things are most stable if they share a supply. These regulators are = nothing=20 fancy, but Good Enough considering the power supply rejection afforded = by the=20 extensive use of CCS loading. They're variations of the classic two = transistor=20 regulator I used for the Heretical and, in fact, I pressed some leftover = Heretical circuit boards into service. Noise is low, stability is high, = and the=20 simplicity is appealing. One might substitute a MOSFET for the pass = transistor,=20 but the cheap plain-vanilla TIP50 works very well indeed.
Figure=20 6: HV Regulators
The regulator for the heaters is shown in = Figure=20 7. As before, this is quite straightforward, an exercise in three pin=20 regulation. But there are a few little twists. First among them is the = seemingly=20 odd choice of using two 6V regulators instead of one 12V regulator. This = choice=20 greatly increases common mode rejection, an Achilles' Heel of most = heater power=20 supplies; a common-mode choke will make things even better, but I just = didn't=20 have one. If you do, feel free to use it.
The more well-known = twist is=20 the use of R30 and R31 to elevate the heaters 65V above ground. This has = two=20 salutary effects - first, it reduces the heater-to-cathode strain of the = cathode=20 follower. For that tube, the cathode is roughly 95 volts above ground. = Without=20 this heater elevation, the heater-to-cathode voltage well exceeds the = 50V limit=20 for triode section one and strains the limits of triode section two. = With 65=20 volts of elevation, the heater-cathode potential for the follower is a = balmy=20 30V.
Figure=20 7: Heater supply regulators
The second salutary effect is = that=20 heater supply noise is less likely to be injected into the first or = second stage=20 cathodes. One thing that helps is that the cathodes are pretty well tied = to AC=20 ground (not degenerated). Nonetheless, noise can couple efficiently = between=20 heater and cathode via a diode-like interaction. Elevating the heaters = seems to=20 ameliorate that coupling. C15 ties the center of it all to AC ground. In = the Raw=20 Supply section, I'll mention one more trick to rid us of heater circuit=20 noise.
In any event, the required heater current is about = 600-700mA=20 (depending on which ECC88 variant is used). With a raw H supply of about = +/-10V,=20 each regulator dissipates about 2.6W, so should be heatsinked.
The External Power Supply
The raw supply is even more = embarrassingly=20 straightforward. It's built into a separate box using only the finest = Radio=20 Shack $10 cabinet (figure 8). I used mostly on-hand and surplus parts, = but have=20 given part numbers for some currently-available units. The HV power = transformer=20 was a lovely surplus item, potted and shielded. The one in the parts = list is=20 amazingly well-priced and should work fine. I'd avoid toroids since they = are=20 superbly efficient at coupling mains noise into the circuit.
A = nice=20 trick- use a separate transformer for the heater supply rather than the = usual=20 extra winding on the HV transformer. The reason for this is that, even = using=20 high speed soft recovery diodes, rectifier hash from the HV supply will = be=20 coupled to the heater winding, providing unwanted rattle. This is a = minor=20 effect, but we're handling microvolt signals=85
If=20
you're extremely lazy (who, me?), there's a nice cheat that works great =
and=20
saves money and effort- there are lots of wall-wart supplies available =
surplus.=20
Find a couple of nice heavy ones (conventional supply rather than =
switching=20
brick) rated at 9V/1A; I dug mine up for $2 each. They can be fit inside =
the=20
power supply box in place of the discrete circuits. Nonetheless, I have=20
indicated more generally available parts, but don't hesitate to visit a =
surplus=20
shop and improvise. Again, laziness compelled me to use a power entry =
module=20
(PEM), a chunk of plastic containing an IEC inlet, power switch, fuse, =
and RF=20
line filter. I dug a pile of them out of a surplus bin for maybe $3 =
each- you=20
should be able to do as well. The earth ground should be securely =
attached to=20
any exposed metal parts (in my case, the top plate and HV transformer).=20
The sharp-eyed might have noticed that the transformer in the = photo only=20 has a 200V secondary. This was boosted to 230V by connecting the mains = to the=20 lowest primary tap. The raw B+ comes in at 285-300V, depending on the=20 transformer, which allows a nice 25-40V cushion for the regulator (the = higher=20 the better).
The schematic for the raw supply is shown in Figure=20 9.
Figure=20 9: The raw supplies, high voltage and heater
Building the Signal Circuit
I freely admit that my first = prototypes are usually a bit=85 rough. This one was no exception. The = next build=20 will be lovelier. But it does work. I found a Collins S-Line-style = cabinet in=20 the back of a surplus shop so grabbed it immediately. The front received = a 5 pin=20 DIN connector for the balanced phono, the rear took a surplus Amphenol = connector=20 for the power and RCA plugs for the single-ended output (Figure 10). =
The=20
internals are shown in Figure 11. Some digging around and filing got me =
a place=20
to mount the perfboard. All signal and ground wiring was done with solid =
silver=20
wire with Teflon sleeving. The schematics give a clue about the =
grounding- I=20
used a combination of star and bus, with all grounds returning to a =
large solder=20
tag right next to the input DIN jack.Since the photos were = taken, I've=20 added heatsinks to the "upper" CCS transistors in the phono stage, = twisted the=20 gray and white transformer secondary leads on the lower right (I think=20 forgetting that was one of my late-night moves), and added a 1k8/5W = resistor in=20 series with the collectors of the pass transistors for the 160V rail. = I've also=20 moved the earth ground lead from the raw supply from the chassis next to = the=20 power input plug (the blue wire) to the single point earth ground to the = chassis=20 next to the DIN input connector. This made a major difference in the = noise=20 floor, with a sharp reduction in odd harmonics of 60 Hz.
Tubes=20
are, sadly, microphonic. This is one case where fanatic antiresonance =
measures=20
and shock mounting really pay off. Me, I just put a piece of foam rubber =
on an=20
old VPI turntable platform and stuck the preamp on there. No huge =
ringing=20
problems, but a nice build will mechanically isolate the signal circuit =
so that=20
I don't hear the constant whining about how trailer-trash the whole =
setup looks.=20
Designer components may be nice, but I just don't see the point. = Use=20 good-quality polypropylene coupling caps, carbon resistors for the gate = and=20 grid-stoppers, and 105=B0C-rated electrolytic caps in the power supply = and=20 regulators. As long as you don't look inside, the sound will be every = bit as=20 good as the fancy spread.
Trimming the RIAA network isn't too = hard.=20 Adjust the buildout resistor R7 to get the 1kHz response to be 20 dB = below the=20 50Hz response. Then adjust C4 to give response at 20kHz at about -19dB = with=20 respect to 1kHz (if you're using the Missing Zero) or -19.6dB if you're=20 not.
The=20 cabling between power supply and signal circuit is likewise = straightforward, but=20 make sure you include a separate earth ground lead between the boxes so = that all=20 exposed metal parts are at safety (third pin) ground!
Proof of the Pudding
A noise spectrum of this box of = gain, with=20 the cartridge connected, is shown in Figure 12. It clearly follows the = RIAA=20 curve indicating that the primary noise source is indeed the cartridge = and first=20 stage. Total noise was measured over the range 20Hz to 20kHz with a 60Hz = notch=20 filter (the hum seems to be a function of how I run the = cartridge-to-preamp=20 interconnect) and found to be (unweighted) 0.16mV, which is very close = to the=20 noise prediction of -68dB with respect to 0.5V.
Figure=20 12: Noise spectrum with cartridge attached
The 60Hz spike = was=20 bothersome and, in fact, when I turned the volume to 11 and stood very = near the=20 speakers, I could barely hear it. But I could hear it=85 This was one of = a=20 continuing set of lessons in Small Things Count- I discovered that I had = neglected to twist together the leads between the input transformer = secondary=20 and the D3a; once that was done, the tiny bit of residual hum = disappeared.=20
The spectrum of a 1kHz tone at 3VRMS output is shown in Figure = 13; this=20 was generated by feeding a signal through the attenuator in Figure 2, = not from=20 the actual cartridge. The second harmonic dominates at -70dB (0.03%). = The THD is=20 a strong function of the tube choice for the second voltage amplifier. = For=20 example, substitution of a 6KN8 raised the distortion to 0.06% (still=20 predominantly second harmonic). An Amperex Bugle Boy clocked in at = slightly=20 under 0.05%, and the champion was the Siemens CCa at 0.03%. Swapping = input tubes=20 made almost no difference.
Figure=20 13. Spectrum of 1kHz tone at 3VRMS out
Speaking of input = tubes,=20 most D3a that I've checked needed some time to burn in before their grid = current=20 stabilized. One way to do this is to bake the tubes as suggested in = "Valve=20 Amplifiers," 3rd edition. I'm a bit leery of that method, worrying about = compromising the glass-to-metal seal where the pins enter the envelope; = my own=20 choice is to burn in the tube in situ, grinning and bearing the = sound=20 until all has reached equilibrium.
Enhancements
This preamplifier has several compromises for=20 practicality. Chief among them is the coupling methods mentioned before- = an=20 all-out version would have a direct coupled output with a servo. I won't = argue=20 with anyone who'd go in that direction.
For more gain, the second = stage=20 tube could also be a D3a. Set its CCS load at 10mA and use a 1.7V red = LED in the=20 cathode. True studs will also use a D3a for the cathode follower- the=20 transconductance and gain give it a theoretical advantage. In the Real = World,=20 the follower's performance is already quite impeccable, so this is = really more=20 for show than for go.
Sonics
I'm not one who is much for the purple prose of audio = writers. Nor did I do a rigorous double-blind level matched test = comparing this=20 preamplifier to my old one. But, my uncontrolled subjective impressions = were=20 consistent with the measurements- quiet, clean, and unobtrusive. The = shattering=20 mistracking and noise on some of my older, lousier records seems to be = much less=20 noticeable. No oddball blats, buzzes, or shrieks. It's really pretty=20 delightful.
Acknowledgments
So many discussions with so many people = about phono=20 stages! But I really should point to John Curl, Morgan Jones, and Allen = Wright=20 (Vacuum State - = High End=20 Hifi Equipment) for many long and involved arguments that really = shaped my=20 thinking here. Many of the critical parts came out of Mr. Jones=92s = Locker. Tim de=20 Paravicini declared this a "terrible" design, and I appreciate his = input. Dave=20 Dlugos (planet_10=20 hifi) scrounged up the Technics MC, an amazing feat (can you find an = EPC100C=20 Mk4 that=92s been in stasis?). Without Cynthia Wenslow, the mechanics of = this=20 article would have been impossible; all the good photos were hers (and = her=20 copyright, used with her kind permission), lousy ones were mine. Jan = Didden (
And as = usual,=20 great discussions amongst the denizens of diyAudio.com were a constant=20 inspiration.
References and Further Reading:
Jim = Hagerman, "On=20 Reference RIAA Networks,," available at
Morgan Jones, "Valve = Amplifiers," 3rd edition, Newnes 2003.
Walt Jung, audioXpress 4/09 = "High=20 Performance Current Regulators Revised"
R. Landee, D. Davis, and A. = Albrecht,=20 "Electronic Designers' Handbook," McGraw-Hill 1957.
Stanley Lipshitz, = "On=20 RIAA Equalization networks," JAES 27:6, 458, 1979.
National = Semiconductor,=20 "Audio Radio Handbook," 1980.
Allen Wright, "The Tube Preamp = Cookbook," 2nd=20 Edition, Vacuum State Electronics, 1997.
Stuart Yaniger, audioXpress = 2/09=20 "The ImPasse Preamplifier"
Stuart Yaniger, "The Heretical = preamplifier,"=20 available at SYclotron Audio The Heretical Preamp
PARTS = LIST:=20 SIGNAL CIRCUIT (one channel, two needed)
C1, C3 0u1 400V =
C2 36n*=20 400V (33n plus trim for RIAA)
C4 2u2 250V
C5 1u 400V
D1 IR = LED=20 (1.25V)
D2 Red LED (1.7V)
D3-D4 LM329
NE Neon bulb (NE-2 or=20 equivalent)
Q1-Q6 DN2540A
R1 6k8 0.5W
R2 100R 0.5W
R3, = R11,=20
R12, R16 1k 0.5W
R4, R6, R13,
R14, R15, R 17,
R18, R19 = 300R=20 0.25W
R5 120R 0.25W
R7 20k* 0.5W (trim for RIAA)
R8 3k3 0.5W =
R9=20 91R 0.5W
R10, R21 1M 0.5W
R20 150R 0.25W
T1 Sowter 8055X, = 1:10=20 step-up transformer
V1 D3a pentode
V2 ECC88 or equivalent dual=20 triode
PARTS LIST: HV REGULATORS (One channel, two=20 needed)
C6, C9 0u22/400V
C7, C10 47u/400V
C8, C11=20 0u1/400V
Q7, Q9 TIP50A
Q8, Q10 MPSA42
R22 51k/2W
R23 = 160k/1W=20
R24, R27 8k2
R25 68k/2W
R26 220k/1W
PARTS = LIST:=20 HEATER REGULATOR (both channels, one needed)
C12, C16=20 0u1/100V
C13, C14,
C17, C18 220u, 25V
D5, D6
D7, D8 = 1N4007
R28,=20 R32 470R
R29, R33 120R
R30 100k/2W
R31 33k/1W
U1 LM317 =
U2=20 LM337
PARTS LIST: RAW SUPPLIES (both channels, one=20 needed)
C19, C20 4700uF/35V electrolytic capacitor
C21 = 47uF 450V=20 electrolytic capacitor
C22 470uF 450V electrolytic capacitor
D9, = D10=20 MUR4100EG fast recovery diodes (4A, 1000V)
D11-D15 5A, 100PIV (or = more)=20 rectifiers or bridge
L1 Choke, 5H/100mA or greater (Triode = Electronics=20 M100D)
R34 470k/2W
T2 Power Transformer, 250-0-250V, 100mA (Edcor=20 XPWR001)
T3 Power transformer, 18-20VCT, 1A (Xicon 41FJ010 or Edcor=20 PWRC20V1A-1)
PEM IEC Power entry module, 3A, with line filter, fuse, = and=20 switch
FIGURE CAPTIONS:
Figure 1. Passive EQ RIAA = stage=20 topology
Figure 2. Attenuator for measurements
Figure 3. First = stage=20 topology
Figure 4. Thermal noise from RIAA network
Figure 5. = Signal=20 circuitry
Figure 6. HV regulators
Figure 7. Heater supply=20 regulators
Figure 8. The external raw supply
Figure 9. The raw = supplies,=20 high voltage and heater
Figure 10. Low-rent casework; a) DIN balanced = in; b)=20 RCA single-ended out
Figure 11: Low-rent innards. See text for = changes since=20 this photo.
Figure 12. Noise spectrum with cartridge = attached.
Figure 13.=20 Spectrum of 1kHz tone at 3VRMS out.
=20
=20
thread;
=20
almost enough xformers , but certainly = not=20 enough shunt regs