Before I started writing Part II, I did a bit of arm-twisting when John got back from Vietnam, hoping he’d do some of the heavy lifting for me and make a series of measurements that would illustrate the points I alluded to in the first part of the Filament Supply article. John, as usual, measured things to a fare-the-well, and gave me more things to write about. With a big hat-tip to John A, all of the figures and test data in this article come from Research Report #002, DC Filament Supply Test.
The usual DC power supply for DHT filaments is similar to a typical transistor-amp power supply, with a high-current transformer driving a four-diode bridge, and the bridge connected to a large-value first cap. (Transistor amps typically have a plus and minus supply, while a filament supply has only one rail, but that doesn’t change the basic behavior of the supply.)
It should be kept in mind the first cap is not a filter; to act as a filter, it would have be preceded by a known resistance or an inductance. As it is, the diode bridge is severely nonlinear, so cannot be analyzed in classic RC or LC filter terms. A more accurate characterization of the first cap is a charge/discharge element; the rectifier bridge is only turned on for the brief moment when the incoming AC voltage is higher than the stored voltage of the DC side of the circuit. At all other times - and certainly through the zero-crossing region - the rectifiers are completely shut off and the circuit is powered from stored energy in the first cap.
What interesting is the consequence of the DCR of the large electrolytic cap, the rectifier bridge, the power transformer, and the power line going into the circuit. The DCR is only thing limiting the current flow; in principle, with zero DCR, the current flow into the first cap would be infinite! Obviously, the DCR isn’t infinite, but it isn’t a controlled variable, either. As the power transformer gets bigger, the diode drop goes lower, and the stray DCR of the first cap goes lower, more peak charging current flows into the first cap, and the charging interval gets shorter.
Circuit #1 implements the usual cap-input filament supply driving a simulated load equivalent to two 300Bs. The current and voltage waveforms are:
So what’s surprising about this scope photo isn’t the 20 peak amps, but that the peak current isn’t even higher, and charge interval isn’t shorter. The incoming AC waveform at John’s house has an unexpected flat-topped characteristic, which limits the peak currents of the test power supply (see postscript at the end of this posting). The rectifier bridge in John’s test circuit is an old (1970) model which probably has higher internal resistance than contemporary ones. Back when I worked at Audionics, we needed a 50-amp bridge to prevent failures in our 70-watt-per-channel amplifiers, and that only used rather modest 20,000uF caps for each rail.
I want the reader to think what this 20 peak amps actually represents: it’s a brief surge of current, a mere 3 milliseconds long, that charges the 20,000uF first cap of the test circuit, 120 times a second. This creates a comb spectra that has significant content across the entire audio spectrum - see the FFT of the charging current for Circuit One:
The current loop (total area of the power cord, transformer, bridge, and first cap) radiates this 3 millisecond magnetic pulse, which spreads out in all directions. This is why techniques like twisting the wire in the power cord, and minimizing the loop-area of the filament power supply, makes a significant difference to the overall noise level - not just to the power amp, but the entire audio system.
That’s the magnetic component: the current pulse itself is also radiated in two directions, back out to the power cord and into the house wiring, and forward into the audio circuit. Note that for a typical transistor power amp, there is no additional filtering before between the first cap and the power transistors, and typically only a single pole of RC filtering to power the voltage-gain stage of the amplifier. Now you know why transistor amps need feedback - to keep the noise down!
In Circuit Three, the current pulse is post-filtered, and John shows the results of splitting the 20,000uF into two sections, with a 10,000uF first cap and a RC filter for the 10,000uF second cap.
This wastes a little power, but then again, a voltage regulator wastes some power too, since they need a few volts of headroom to operate. (A minor point - it’s a bit easier to get rid of heat from a resistor, since it won’t require the heat-sink, grease, and insulated mounting pad that a transistor regulator requires for reliable operation.) Notice how even a simple RC post-filter reduces the noise quite substantially - although the rest of the amplifier still has to contend with the magnetic field from the current pulses, since there’s still 20 amps going into the charge/discharge first cap.
Circuit #2 shows a different, and smarter, way to approach the problem: limiting the rate-of-charge of the first cap by the simple expedient of intentionally raising the DCR of the whole circuit (rather than let it be controlled by strays in the cap, power transformer, and power cord). This drops the peak current drawn by the first cap down to 9 amps, which then in turn drops the overall noise level of the entire circuit.
Circuit Two and Three could be combined, at the expense of a bit more power loss, so there would be four resistors in all: a pair on the high and low-side between the bridge and first cap, and a second pair on the high and low-side between the first and second cap. This would approach the noise levels of Circuit Four, but without the weight and cost of the choke filter. It would also have a minor advantage of a slow-start for the filament, since the cold resistance of a filament is about 1/5 of the hot value, and the noise-filter resistors would act as current limiters until the filament reached operating temperature.
Circuit Four shows the classic solution for low noise: a true choke-filter power supply. This has the merit of constant current draw from the AC line, no sharp switching waveforms, and the ability to use DHT’s with different current draws without re-designing the circuit, due to the improved regulation of a choke-input supply.
The comparison between the voltage FFT’s are illuminating: look at the difference between Circuit One and Circuit Four at 3kHz, where the ear has the greatest sensitivity to noise and distortion. There’s an astonishing 55 dB difference between the two circuits!
Now, maybe you have a lot of faith in an expensive active voltage regulator, but I wonder if it would really deliver a 55 dB improvement at 3kHz? Remember, regulators use feedback to do their magic, and the feedback is most effective at DC, and decreases at 6 dB/octave as frequency goes up. By contrast, the choke filter gets better as frequency increases, up to the point where stray capacitance makes the choke resonant.
Considering that elsewhere in a power amp we’d be struggling for a 6 dB improvement in noise (especially buzz and hash levels), hand-selecting tubes, careful wiring layout, etc. etc., how can anyone dismiss a 55 dB improvement?
If you were really hard-core, there would be four chokes: a pair on the high and low-side between the bridge and first cap, and a second pair on the high and low-side between the first and second cap. The first pair, since it emits magnetic noise, would be physically isolated from the audio circuit (on the far side of the chassis, close to the dedicated filament transformer), and the second pair would be close to the DHT, since it is part of a LC filter circuit. Another improvement would be to insert an RF common-mode choke of the type seen in the input of computer power supplies. This would reduce the common-mode RF noise that can sneak through the large power-frequency chokes.
Postscript (by John Atwood)
The sluggish charging pulses with the cap-input supplies surprised me, since I was used to seeing narrower, taller pulses, just as Lynn mentioned. I looked at the waveform of the power line, measured at the secondary of the power transformer with no load. Just to check that this wasn’t due to a bad outlet or unusual load, I took a portable scope around the house, and indeed this is the waveform of the power coming to my house:
It used to look more like a sine wave when I first moved here (to Nevada) eight years ago. My house was built in 1977 and has 200 Amp service, higher than the average house. However, I live at the tail-end of a rural electric distribution system. The “flat-topping” is likely caused by an excessive amount of cap-input supplies on the line, perhaps due to the increasing amount of computers and electronic ballasts, which often use cap-input supplies. Unlike Europe, America doesn’t have regulations on the power factor of devices connected to the power lines, so manufacturers of devices with switching power supplies nearly always use cap-input rectifiers, right off the power line.
Where is the test mesurement of FFT’s for circuit #2 and #3?
I am doing a similar circuit like #3 for 6.3 volts @ 1 amp
for a 6B4G amp but I have yet to get all the parts in to build the circuit.
One advantage of circuit #3 is that you have some inrush current limiting
on the DHT’s and that may help them last longer.
5 watt 3.3 volt zener diodes may be useful for a simple zener
pre-regulator for a more complex design for a single 300B.
Woodelf: I only did the FFTs for cases #1 and #4. The current noise on #3 will be similar to #1 and the voltage noise on #3 will be similar to #4. However, I can resurrect the set-up and run the FFTs. Give me a day or two. Any series resistance will provide inrush current limiting which will slow the warm-up of the tubes. The choke-input filter is the worst in this regard, although extra resistance could always be added.
John,
if you’re going to resurrect it, perhaps it would be nice to see what the effect of schottky or soft switching diodes might be. Or the effect of a small snubber cap?
John,
Interesting results. Your comments about how well a voltage reg would fare got me thinking, as I use a three pin reg wired as a current reg. So I tried a couple of tests of my own. First I looked at the spectra of a circuit similar you what you were using, 9v toroid, Skot bridge, 10,000uf cap, resistor and load. This is what normally feed the regulator. So I to a look at the harmonics from that point, and using the same settings on the analyser, replaced the dropper resistor with the current reg.
The results can be found in these scans
http://home.lurcher.org/nick/audio/FillSpect.pdf
Simple answer to the can it give 55db of reduction, answer, maybe, as it hits the noise floor, but it seems nice and quiet (more so than I expected in fact).
I hope you get the chance to repeat this and see how it looks to you.
NickG - I’m sorry, your comment was stuck in moderation for some reason, and I only noticed it while clearing out spam. I’ve been tightening the spam filtering and you got caught (but I didn’t see any “bad” words ???).
Quite impressive curves! The active current regulator certainly does a good job. What is the DC drop across the regulator?
An active current regulator is doing what an inductor does - try to keep the current constant, which ideally will eliminate ripple. The inductor has an advantage over an active regulator - it is an energy-storage device, so with an ideal inductor, there doesn’t have to be any DC voltage drop across it for it to work. An active regulator doesn’t store energy, so it must have a DC voltage drop greater than the maximum ripple for it to work. Of course, inductors are far from perfect, with winding resistance, weight, cost, and a nasty magnetic field.
I’m a bit of a traditionalist, so prefer to use classic devices where I can. Despite (or maybe because of) working in several semiconductor companies, I like the old-fashioned way, so will probably stick to my inductors where they make sense. I’m not against technology where it’s needed, though.
By the way, how do you like your Spectra-Plus FFT system? I always get asked for good FFT software, and am not familiar with this one.
- John
NickG,
I have a question regarding the spectra: In the header it says Cap”filter”, 120mV RMS and for the other Current regulator, 15mV RMS. I assume these are the results for the total noise. The difference is about 20dB. If I look at the spectra, the difference seems to be much larger (70dB for the 100Hz peak).
How can this be?
I performed a series of differential noise measurements on DC filament supplies. The results are quite different from normal (grounded) measurements. A VCCS is not at all that good measured this way.
Killer combination:
Switching power supply followed by a VCCS. Total ripple & noise = 0.1mV (-79dbV). The spectrum is flat, yes flat, from 20Hz to 20KHz.
Click on my bame below the post, and look for Audio Projects > DC Heater supply part 3, the final insult.
Mea Culpa…
I commited a terrible gaffe, turns out that the switching power supply with the excellent performance had an output connected to ground. I tried a floating switching power supply and it performed a lot worse.
So I changed the way I measured. The dummy load resistor (3.3Ω for 1.5A at 5V) is now connected to ground just if it were in a single ended amp, with a 1K cathode resistor decoupled with a 100µF cap.
This yields yet another set of results. Most interesting was that I could now measure the signal across my ‘cathode resistor’, this signal turns out to be dependend on the amount of common-mode noise on the heater supply.
The best performance I achieved with the following circuit (I had no suitable choke to test):
- 12 AC from the Lundahl 1651
- A rectifier bridge made up of HEXFREDs equipped with 22nF snubber caps
- A 0.5Ω resistor to reduce the peak currents and drop the input voltage for the VCCS to 9.3V
- A symmetrical CRC filter made up of a 6800µF cap 2 X 1Ω resistors and a 15000µF cap.
- A VCCS module (by MachMat)
The differential noise on the dummy load is now 0.062mV (noise level for the measurement = 0.009mV). The differential noise on the ‘cathode resistor’ is 0.441mV (measurement noise = 0.012mV).
So I lowered the value of the cathode resistor to 500Ω, the noise level did not change. Lowering the value to 21Ω dropped the noise to 0.256mV. Lowering the value to 1Ω (a fixed bias configuration) resulted in 0.017mV noise, just above the detection limit.
Questions I have:
- Is it not so that it is the current in the output circuit that matters?
- Lowering the value for the cathode resistor increases the current to ground, does this mean that the common-mode noise has a fairly low impedance?
- What does this all mean for the fixed vs autobias discussion?
Gerrit - Sorry I haven’t had a chance to comment sooner, but I have been driving across the US, and tomorrow should arrive in Montreal for the Festival Son & Image audio show. I’ve got a few hours now, and had a chance to look over your write-up on your web-site. Quite a good job! Here are some comments on the filament supply problem in general, then I will get to your specific questions.
The common-mode noise, i.e. the noise between the combined power supply outputs and signal ground is mainly caused by capacitive coupling between the raw power line and the rectification/filtering part of the supply. In a conventional supply, this is caused by capacitive coupling through the power transformer. The use of an electrostatic shield, connected to the signal ground, can help reduce this. A switching power supply has a much smaller transformer, and is usually wound so that there is good separation between the primary and secondary, so it is not surprising that the common-mode noise is lower.
Some of your noise measurements are well below 1mV, so extra special care is needed when making these measurements so that extraneous noise is not introduced. Your measurement set-up should have floating inputs with good common-mode rejection. Be careful not to have open loops in your connections that could pick up magnetic hum fields. Doing a “zero-signal” run, like you did, is a good way to see if you have noise problems in your set-up.
The 100uF bypass capacitor has a capacitive reactance of about 32 ohms at 50Hz - a little borderline, IMHO. This is why the differential noise didn’t drop until the cathode resistor was in the tens of ohms - about the same as the capacitor reactance. I would use a higher value cap - maybe 500uF or 1000uF, although it may be harder to find a high-quality cap at these values.
The “current in the output circuit” (I presume the plate current) shouldn’t matter, but the cathode resistor does, and this indirectly is affected by the current. However, if a perfect bypass capacitor is used, there should be no noise. Since nothing is perfect, there is some compromise here. This is one reason why fixed bias (with little or no cathode resistance and no bypass cap) is better. (The other reason is that with cathode bias, the bias can slowly move around when loud transients or overloads occur, giving a “woozy”, less stable sound.)
The common mode noise from a purely floating supply should be a capacitance that is not very high, thus the impedance is high. However, the interfering signal can be high, too, so a significant noise current can be injected. If there is a common ground connection, say, through the power cord safety ground, the interfering common-mode signal can be high and a low-impedance. Since you are doing FFT spectrum plots, you can get an idea of the nature of the noise coupling and its filtering. If you get a monotonic decline in line harmonics with increasing frequency, then the spectrum is caused by a dominant bypass cap, such as your 100uF bypass. If, on the other hand, the noise is flat or even increasing with frequency, then the noise is being dominated by the series coupling (possibly through the power transformer) capacitor.
I hope this makes sense. It is a complex topic, but we are making stabs at clarifying it. Keep asking questions, and I will try to answer when I can.
- John
John,
Thanks for the compliment, and the thourough response. Your comments on the bypass cap prompted me to do some testing, I found that increasing the value results in an almost linear reduction of the noise. But as you said, large value bypass caps are not easy to find.
In the meantime I finished the amps and I’m very pleased with the result. Soundwise it is already a serious improvement over my existing amp.
There is no ’startup noise’, it is absolutely dead quiet. Putting all this effort in the power supply surely paid off, there is no hum, no hiss, no noise, nothing. Measured noise on the output is 55µV or -85dBV. It only makes some funny noise when you turn it off.
The pictures and graphs are on my website.