I use both depending on whether or not the use of negative grid bias requires the use of an additional coupling capacitor - in which case I will probably use the LED instead. If the coupling capacitor is there already, and I can conveniently derive the required grid bias then I will go that route. In the event where I need the lowest possible rp to drive a choke or transformer with a high mu/high transconductance triode then I will probably select fixed bias using a negative grid supply (or battery) and live with a high quality coupling cap. With choke loading at least I have not measured a significant degradation in linearity - at least one I can reliably measure. (Perhaps a dB or so..)
Guess all of this makes me pretty agnostic on the question of LED bias, I use whatever suits the topology and maintains an acceptable level of simplicity. That said I have not heard nor measured anything that leads me to believe the LED is a bad idea. (Actually pretty much the contrary since I first started using them a couple of years ago - yes I was a "late" adapter..
)Unbypassed cathode resistors generate noise which is gained up by mu+1, and I have not found any (even quite exotic) electrolytics that perform in a fully satisfactory manner with high transconductance (low effective -Vg) triodes or triode strapped pentodes which require very large values of capacitance for a reasonable LF corner. A properly chosen LED seems to be greatly superior to the traditional RC derived cathode bias under these conditions particularly if the tube in question is loaded by a CCS or gyrator.
Above equation applies to a regular PN junction diode. You can rearrange the equation to show that if you keep the current constant, the voltage drop across the diode will remain constant. That's the whole premise of the CCS+LED biasing scheme.
In reality, the current through the diode will vary a tiny bit and this will cause the bias point to shift however slightly. To the first order, this means the diode can be modeled as a constant voltage with a small series resistor. I've measured a couple of LEDs on a curve tracer. The ones I found (red LED, forget part number) could be approximated by 1.8 V, 20 ohm for any current in the range of (roughly) 500 uA ~ 30 mA. I didn't measure past 30 mA... The dynamic resistance changes very slightly depending on the DC current through the diode and this will cause some distortion if the DC current varies as function of input signal level. This is why it's important to use a quality current source.
The bottom line is that all practical circuit components will introduce some amount of non-linear distortion. No component is ideal.
~Tom
That's unusually high resistance. Far more typical values are 1R-5R.
Let's look at scale. The source impedance of a cascode CCS is likely to be 10M or higher; Jung's measurements of the MOSFET cascode that I use show 100M+. For a 20V swing at the plate, then, delta I = 0.2uA. For a 5R dynamic resistance LED, that's an AC voltage of 0.2uV across the LED. For that 20V swing in a high mu tube like a 12AX7, the input voltage is 200mV. So the impressed AC voltage (not the distortion!) from the LED is 1ppm or 0.0001%. So even if the nonlinearity of the LED's dynamic resistance were 100% (it's nothing even vaguely close to that), the contribution of the LED is -120dB! For a more realistic nonlinearity, we're talking more than 140dB down, or 1/20,000 of the distortion contribution of the tube. How good is a negative bias supply?
I suspect this is not a serious problem.
Like tomchr I have also run into some red leds that had high dynamic impedance as well, in the range of 10 - 15 ohms, however they are more than 20yrs old, and given changes in process and doping I suspect they are a lot worse than their modern cousins in this regard. Note that they still work quite a lot better than the RC based cathode bias circuits they replaced.
That's unusually high resistance. Far more typical values are 1R-5R.
Let's look at scale. The source impedance of a cascode CCS is likely to be 10M or higher; Jung's measurements of the MOSFET cascode that I use show 100M+. For a 20V swing at the plate, then, delta I = 0.2uA. For a 5R dynamic resistance LED, that's an AC voltage of 0.2uV across the LED. For that 20V swing in a high mu tube like a 12AX7, the input voltage is 200mV. So the impressed AC voltage (not the distortion!) from the LED is 1ppm or 0.0001%. So even if the nonlinearity of the LED's dynamic resistance were 100% (it's nothing even vaguely close to that), the contribution of the LED is -120dB! For a more realistic nonlinearity, we're talking more than 140dB down, or 1/20,000 of the distortion contribution of the tube. How good is a negative bias supply?
I suspect this is not a serious problem.
Last edited:
This aplies to any diode, Si, Ge, SiC, SiGe, GaAs, GaN, GaAsN and whatever. The only difference is Is. LEDs are as well regular PN diodes. The only different between LEDs and Si diodes is, that Si is an indirect semiconductor and GaAs (low-cost red LEDs) is a direct one.
The 10 ohms (or somthin in this range) you measure is the intrinsic bulk resistance of the LED. In this way you have a series resistance Rs with a pn junction.
You can modify the equation i = is * [exp (U / Ut) - 1] slightly and you get
i = is * [exp ({U - i * Rs} / Ut) - 1] (be happy to solve this equation
)For currents where the dynamic impedance is low against the bulk resistance (at about 10-20 ma) the total resistance is nearly constant and you can model the LED as a constant voltage source in series with a resistor. Unfortunately most tubes are not driven with 10-20ma.
But you're right, if there's no delta I the resistance of a pn junction diode is constant. The question is how to achive this...
I don't see a CCS load, however ideal, being an effective palliative for the dynamic current changes from the tube's Miller capacitance and the load impedance that can also cause nonlinear behavior by the LED (primarily expressed as current dependent changes in its forward voltage drop). Perhaps bypassing the LED with a capacitor would be of some assistance here
Last edited:
Perhaps bypassing the LED with a capacitor would be of some assistance here
I did that the other day (red LED + tantalum electrolytic, my PCB layout called for a device with 100 mils pin spacing which was initially meant to be a vertically placed resistor) and couldn't hear any difference compared to LED with no bypass.
I've gotten most of the advantage of an electrolytic capacitor paralleled with the cathode resistor by merely using a lower value resistor to ground and pulling up to the supply rail with a high value resistor. For a 12AX7, instead of using, say, a 1K Ohm resistor from the cathode to ground, I was able to use ~150 ohms with a suitable pullup to the plate supply. Sure, I wasted a few milliamps of current in this case, but I also obtained most of the noise figure improvement obtainable with a paralleled capacitor without any of its potential degradation. This technique might be helpful in linearizing LED bias also
Which are all PN junctions. I guess I have to be more careful with my choice of words. Apparently my use of the word 'regular' was interpreted differently than I intended.
And your point?
I'm running 5842, 6SN7/6J5, 6BX7 at those currents in a 300B SET amplifier I'm building at the moment.
Achieving zero delta_I is not possible. But making delta_I very small to the point where the LED causes less distortion than a resistor-cap combo or other biasing scheme is possible. And that's really the fundamental point of this discussion.
I mucked with this a couple of weeks ago. Setup as follows: 6J5, common cathode stage, grid grounded through 470 kOhm. It was loaded by roughly 150 kOhm (cap coupled) on the output. B+ was 375 V.
Plate load - bias network (cathode to gnd) --> Results
-------------
12 mA CCS - 2 x red LED --> Gain: 20 V/V, THD: 0.244 %
22kOhm - 2x red LED --> Gain: 15 V/V, THD: 0.68 %
22kOhm - 330R --> Gain: 11.6 V/V, THD: 0.384 %
22kOhm - 330R||1200uF --> Gain: 15 V/V, THD: 0.682 %
12 mA CCS - 330R||1200uF --> Gain: 20 V/V, THD: 0.262 %
The input signal was 1 kHz @ 1 V RMS in all cases. The 22 kOhm plate load resulted in approx 10 mA of anode current.
I later ran with 12 mA CCS, 330R and preferred the sound of this over the 330R||1200uF, though, I did not capture the THD for that combo.
The CCS was an IXYS 10M45. You might be able to do better with a cascoded CCS, but I'll leave it up to you to figure out if it's worth the hassle or not.
~Tom
Last edited:
Tom your measurements are much more comprehensive than those I have made, and clearly does demonstrate the inherent "goodness" (equivalency) of the LED as compared to an RC cathode bias network from a linearity standpoint.
From my own experience with phono stage design using very high transconductance tubes operating at very high currents there are problems with linearity in very large cathode capacitors. The capacitance values required (1000uF) with small resistors for a corner below 5Hz push the result very strongly in favor of the LED were the differences are not at all subtle. You would use such a low or lower corner in the first stage of a passively equalized phono stage in order not to compromise the low end response of the equalizer. Large electrolytics do not behave linearly with a low polarizing voltage (~ 1.2V here) and very small audio signals - generating massive amounts of distortion and ringing across the cap on any signal approximating a musical waveform. The input stage was a triode connected D3A running at almost 20mA with a 62 ohm cathode resistor and a 1000uF cap for a - 3dB point of roughly 2.5Hz, it sounded terrible! I tried Nichicon Muse which was absolutely terrible, and then Blackgate STD which was a bit better but not good enough. Adding a bunch of smaller electrolytics and a film cap in parallel helped significantly, but still did not sound quite right. Static sine wave signals did not reveal the problem, but music and gated bursts did. With the IR LED I selected there is no significant distortion levels measurable at the signal levels encountered in this phono stage, and the LF corner is established by the coupling cap between the first and second stages. Sounds better in every way.
Last edited:
I think, for me, it's a toss-up between the LED and resistor bias. Adding a cap across the resistor made it sound worse. I used a Elna 1200 uF/63 V I had laying around. It's a pretty nice cap with low ESR. But I'm sure it still suffers from the normal "electrolytic stuff" (high-ish loss tangent, dielectric absorption, etc.) I agree that setting LF poles at low-impedance nodes is usually a recipe for trouble. In theory, it's easy, but in real life it becomes a headache if you want it to be clean.
With resistive plate load, the gain of the stage will be impacted quite significantly by any resistance in the cathode. An LED is nice if you need the high gain. With the CCS plate load, the cathode resistor should have much less impact as the output impedance of the CCS is quite high.
I hear using a resistor in the cathode rather than an LED may be a bit easier on the tube as it ages, but I don't recall the thinking behind that statement.
Did you happen to measure the dynamic resistance of that LED before you installed it?
~Tom
The problem with the resistor is you get degenerative feedback and it isn't even constant, it varies with the music signal. I never liked the sound let alone the RMAA results. May be OK for very small signals. I hear a huge increase in SQ with an LED vs resistor and for some reason a CCS on the plate with a LED on the cathode sounds better than the sum of the individual parts (there must be some sort of synergy that I don't understand.)
Only scenario I can think of is possibly worn out or noisy/microphonic tubes, they do sound better with a resistor vs LED.
The LEDs that I've used have had such relatively small resistances that I've seen no need to bypass them.
The main function of a bypass cap across a LED is psychological.
Local degeneration (e.g., unbypassed cathode resistor) is a useful tool, but not a universal one. Besides gain reduction, it increases the effective plate resistance and because the feedback factor isn't very large, it will tend to reduce harmonic distortion at the expense of skewing the harmonic order upwards. The well-known Baxandall analysis applies here in spades.
Local degeneration (e.g., unbypassed cathode resistor) is a useful tool, but not a universal one. Besides gain reduction, it increases the effective plate resistance and because the feedback factor isn't very large, it will tend to reduce harmonic distortion at the expense of skewing the harmonic order upwards. The well-known Baxandall analysis applies here in spades.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.