Remember: Tubes circuits use lethal high voltages. Play with them at your own risk!
The C4S stands for "Camille Cascode Constant Current Source." It was designed by Buddha (John Camille). This current source is simple, robust and in my opinion functions very well. Bottlehead sells the C4S as a kit. I have a web page about getting more out of the CCS called Extended Range Use of the C4S.
A constant current source (CCS) keeps the current constant no matter how much the voltage across it changes. The load, meaning both the tube and B+, sets the voltage across the CCS. This means an ideal CCS set to exactly 10 mA will deliver exactly 10 mA with 10V across it, with 300V across it and with 160V dc and 280V pp ac across it.
What does this mean to the audiophile? With a CCS load, the tube now sees the easiest load possible and almost always sounds better and cleaner.
The CCS has a maximum and minimum voltage across it
that is needed for the CCS to operate correctly. At the maximum voltage,
the CCS fails/ dies/ makes smoke etc. At the minimum voltage, the CCS
can no longer hold the current constant, but nothing bad happens other
than it no longer sounds good.
When a CCS exceeds its maximum voltage, it usually fails short. Remember, silicon (FET or BJT) does not have a sense of humor about its rated voltage. When CCSs use silicon for the gain stages, they can be delicate when it comes to voltage. FETs are better about voltage stress than BJTs. You can buy avalanche rated FETs, but not avalanche rated BJTs. An avalanche rated FET means near its rated voltage, the FET will behave like a zener for a brief period of time without damage.
If FETs more robust, why use a BJT?When they say 300V on a Silicon based CCS, the CCS can fail as low as 301V. Most BJTs have a small guard band around their failure point. But there is a "spread" around this guard band. You can happily go building hundreds of 300V CCS that see 320V at power up and/ or at full volume with no failures then all of a sudden, every single one dies at 301V. A tube CCS usually has a sense of humor about it's operating voltage because a tube is usually rated at it's nominal bias point, not its peak voltage. Silicon CCS are rated at their failure points.
BJTs are less static sensitive and usually have a lower (better) output capacitance than FETs.
Why use silicon based CCS instead of tubes?BJTs have an accumulative failure mechanism where if you exceed its rated voltage or its SOA (Safe operating area) by just a little bit, the BJT degrades, but doesn't fail right away. It will build up degradation every time you stress it and because of Murphy, waits until you are showing off your system to fail. This isn't just a text book problem. I learned about this the hard way at my first job out of college. And yep, it waited until the boss was looking over my shoulder to make smoke.
A solid state CCS doesn't need a floating filament supply and if designed with electrical margins, never wears out.Why use a Tube CCS?
The tube CCS will generally have lower output capacitance than a solid state CCS.
There are two minimum voltages to be concerned about. The first is hard to calculate and specify correctly. It is where the CCS's performance starts to degrade. By degrade I mean the capacitance goes up, the effective resistance goes down and the distortion may go up. For high voltage transistors (BJTs) and most FETs, this is 6-10V. For low voltage BJT, it can be as low as 1V. The second minimum voltage is where the CCS "clips." This voltage is easy to calculate, marketing isn't involved and is normally what is specified for the CCS as its minimum voltage.
Lets say we want to drive a 6SN7GT at 10 mA plate current with 195V plate to cathode from a 300V B+. We have the following CCS
Set point: 10 mAWe're are pushing things by running the CCS from a 300V B+, but 300V is at high line and we'll willing to risk it.
Max. rated Voltage: 300V
Drop out voltage: 10V
We'll need the curves for the 6SN7GT so we'll look them up on TDSL found on http://tdsl.duncanamps.com. From TDSL we find the curves we need located at http://www.triodeel.com/6sn7_p4.gif. Once we have the curve, we follow the 10 mA line to 195V and see that the grid voltage is -5V. From this information we can calculate the cathode resistor
R_cathode = V_grid / I_plateI'd round up to 511 ohm (mostly because I have hundreds of them in my junk drawer) so we'll have 10 mA * 511 ohms = 5.11V on the grid. Going back to the tube curves we see we end up with about 200V plate to cathode. This is close enough.
R_cathode = 5V/ 10 mA = 500 ohms.
Now calculate the voltage across the CCS
300V (B+) - 200V (plate to cathode) - 5V (cathode to ground )= 95V.Into an open circuit, the maximum output swing will be smaller of when the CCS clips or the tube clips.
CCS clipping: 95V nominal across CCS - 10V (CCS drop out voltage) = 85V positiveInto a load, the maximum output swing will be lower yet. To figure this out, we'll have to draw a load line. Load lines aren't covered in CCS 201.
Tube Clipping: 200V nominal across tube - 90V (10 mA plate, 0V grid) = 110V negative.
10 mA from CCS doesn't mean the tube will always be conducting 10 mA. The CCS current drives both the tube and the load coupled to the tube. When 100V peak ac is on the plate of a tube driven by a 10 mA CCS and the load on the tube is 100K:
10 mA will be coming out of the CCS,
1 mA will be flowing into the 100K load and
9 mA will be flowing into the tube.
A) Output impedance:
Higher is better, but 10 times the actual load on tube is probably good enough for loading aspects, but more than 10 times can be desirable to support a good PSRR.B) Output capacitance:
Lower is better. At 20 kHz Xc >= 10 times the actual load on tube is probably good enough for loading aspects, but more than 10 times can be desirable to support a good PSRR.C) Slew Rate:Xc = 1/(2 * pi * f * C)in the 100K load example:
C = 1/(2 * pi * f * C)C = 1/(2 * pi * [10 * 20 kHz] * 100 Kohm) = 7.9 pF as a goal. 16 pf will work too. 79 pF is usually trouble.
This requirement is important, but ill defined. Typically, solid state amps say you'll need a slew rate that will support an undistorted maximum output swing at about 200 kHz to avoid sonic pitfalls. Can you use something that doesn't meet this criteria, yes, but it takes more work.D) Stability:A simple discrete CCS with reasonably high bias current through the reference tends not to have problems with slew rate. Reasonably high bias current in a LED reference means the LED is bias greater than 1 mA + CCS current/10. This corresponds to at least 2 mA for a 10 mA CCS.
Higher performance CCS usually have more feedback or use RF parts that can go unstable. Proper grid/ base/ gate stopper resistors will be needed.E) Voltage ratings:
Discussed in CCS 201.1 and CCS 201.2F) Power Rating:
Beware the rated power of transistors. A 100W transistor is rated for 100W when attached to a water cooled heatsink that is held at 25C. You are often lucky to dissipate 10W in a 100W part.G) PSRR:
(Power supply rejection ratio) This is how much the current changes with power supply ripple. PSRR includes more than just the output impedance and output capacitance, it also involves changes in the reference voltage for the CCS.H) Soft Start:
A CCS that slowly comes up to full current is often more desirable than one that slams on. Note: Soft Starts can be bad for CCS's who run from a B+ greater than their rated operating voltage.
I) Distortion in uA or dBuA, (not THD):
Distortion needs to be measured in uA, not THD. You'll probably never see this number quoted just as you never see the current THD of a choke quoted. dBuA is better in my opinion, but some people have problems converting dBuA into uA.
Note: dBuA = 20 * log( amps / 1 uA). 0 dB uA = 1 uA. 20 dBuA = 10 uA. -20 dBuA = 0.1 uA.Why uA? Consider two CCS with 100V ac at 1 kHz across them driving a tube with a 7K plate resistance. The first CCS has a 1000 megohm dynamic impedance with a 30% THD. The second CCS has a 10 megohm dynamic impedance with 1% THD.
The first CCS will draw 100V/ 1000 meg = 0.100 uA of total current and about 28% will be distortion or 0.028 uA.
The second CCS will draw 100V/10 meg = 10 uA of total current and about 1% will be distortion or 1 uA.When the distortion current flows into the 7K plate resistance of the tube, the 1000 meg CCS - tube combination generates 7K * 0.028 uA = 196 uV of distortion and the 10 meg CCS - tube combination generates 7,000 uV of distortion. The CCS with the "higher THD" has 31 dB less distortion when used in the circuit.
If uA (or even dBuA) of distortion were specified, this would be more apparent. The 1000 meg CCS would be specified as -14 dBuA at 100V 1 kHz and the 10 meg CCS would be specified as 16.9 dB uA at 100V 1 kHz or a 31 dB difference between the two. There is a point of diminishing returns. 7 mV -83 dB with respect to 100V. 196 uV is -114 dB down. 7 mV out of 100V is probably adequate for all but the platinum eared Audiophiles.
Besides the maximum and minimum voltage across the CCS, we have to manage the temperature rise of the CCS. Depending on the application, I'll allow a different temperature rises. For a generic discussion of a silicon CCS we'd like to see about 30C junction rise and keep the junction temperature below manufacture's limits into a fault.
So our CCS with 95V across it at 10 mA will be dissipating 95V * 10 mA = 0.95W nominally. For a 30C rise we need
Rja normal operating = 30C / 0.95W = 31 C/W junction to ambient.This means Rtheta junction to case (from the transistor) + Rtheta case to sink (from the mounting insulator/ thermal grease) + Rtheta sink to ambient (from the heatsink itself) has to total less than 31 C/W.
Into a shorted or miswired tube, we'll have [ 300V - 5V (cathode resistor) ] * 10 mA = 2.95W. With a 40C ambient and a 150C rated junction, we are allowed a Rtheta of
Rja fault = (150C - 40C ambient) / 2.95W = 37 C/Wso our nominal heatsinking will be good enough under fault conditions.
This is too complex to discuss here at this time. Briefly, SOA takes into account that the transistor may not be able to support full rated power at all voltages across the transistor.
Recent technical articles indicate that Planar FETs are more suited to amplifier operation than Trench FETs. Trench FETs are optimized for switching, not for amplifier operation. Trench FETs may have to be derated 10 times harder than Planar FETs for reliable operation.
It is easy for the tube characteristics to change more than 10%. These changes will causing the tubes bias point to change. The bias point for a CCS fed tube is varies more with tube changes than with a resistively biased tube.
A CCS biased tube varies more than a resistively biased tube because a tube with a CCS on the plate has a cathode voltage that is effectively a voltage source. A constant current into a constant resistance gives a constant voltage. In a resistively biased tube, if the plate voltage wants to go up, the current into the tube will decrease which in turn causes the cathode voltage to decrease. Decreasing the cathode voltage makes the tube want to draw more current which pulls the plate voltage back down. This "feedback" action stabilizes the bias point. With a CCS on the plate, the cathode voltage is fixed so if the plate wants to be higher at DC, the feedback action to pull it back down is missing.
I've made CCSs that looked like resistors at DC and
a CCS at AC. They were a little "tweaky" they behaved like a very high
value inductor and could motor boat under some transient conditions. To
make them behave, I had to give decrease the AC midband performance a
bit. On the good side, they biased up nicely from tube to tube.
The 115V wall socket voltage can vary 87V up to 132V. 87V is a bit extreme. I've seen 104V and 130V so I'm going to use 132 as high line and 104 as low line or 117V +/- 12% for line voltage changes. In the CCS 201 biasing example, the 300V B+ is hopefully 300V at +12% high line (132V). At -12% low line (104V), B+ will drop from 300V to 236V (64V less). This 64V drop in B+ will decrease the maximum output swing from 85V positive to 21V positive. This may not be enough voltage swing for the amplifier to amplify correctly.
If we throw in a 10% bias point variation (20V) from the tube, the plate can only swing 1V positive at low line when the tube's plate voltage runs high.
This can be fixed by
1. Ignoring the problem and tuning the B+ and CCS for just your house and for the winter/ summer season you'll be using the amp the most. (not the best idea.)2. Adding a feedback loop (Ouch!) or
3. Adding an adjustment pot to the either the CCS current or cathode resistor. Adjustable resistors generally never perform sonically as well as fixed resistors. If we put the adjustment pot in the CCS, the pot won't see the audio current so we have a mild sonic advantage in locating the pot in the CCS. This does present the problem in that the adjustment pot will have lethal voltages on it. If we put the CCS in the cathode of the tube, the pot will be safer to adjust, but will be in the audio path.
4. Making the adjustment by actually changing resistor values in the cathode or CCS.
5. Using something safe to adjust the line voltage to nominal.
6. Recentering the operating point. If the operating point can be recentered, this is probably the best option.
In 301.7 we found our 300V B+ can drop to 236V and we run out of operating margin. When we take off 10V for the CCS our peak output voltage is 226V. This peak is the same no matter what our load line is. This peak is set by B+ and the drop out voltage of the CCS, not the tube.
Looking at the curves for the 6SN6GT at 10 mA. Zero volts on the grid at 10 mA corresponds to 90V. Our best peak to peak output voltage into a very high impedance is
226V (B+) - 90V (plate-cathode) - 5V (cathode to ground) = 131V.If we knock off 10% (or more) for tube variation, the available peak to peak swing becomes 118V or 59V peak. If this is enough. We can continue just by setting the voltage bias point to
90V (plate-cathode) + 5V (cathode to ground) is the bottom of our swing at the plate (95V)We'll do this by changing the grid-cathode voltage to about 3.1V or 3.1V/10 mA = 310 ohms.
226V is the top of our swing at the plate.
95V + (226-95V)/2 = 160.5V is the middle of our swing at the plate.
160.5V - 5V (cathode to ground) is 155V plate to cathode for the tube's bias point at 10 mA
If the load is 51K, the tube will need to conduct 1.2 mA more at 0V grid which takes us up to 100V on the plate.
Just for grins, lets say 61V peak was 9V short of
what we needed. How do we fix this? We need 18V pp more, 20V with 10%
tube margin. -- I'd consider just dropping the current a little bit so
the tube runs at a lower voltage.--
NOT DONE
First edition 10/01/06 Last update 10/01/06
I don't change the update date on individual sections
for minor corrections.
I only change the date for content changes.
(
New 2024 index page.)
_( Old 2003 index page.)
_( AMP Second index
page.)
( Fancy index page.)
