Looking through a lot of relatively new circuit designs of valve amplifiers, the Constant Current Source takes almost all the leading position. Except those used in Filament CCS, almost all other designs are Cascode ones.
For example like this, the M3 and J1-J4 formed a Cascode CCS.
If we set it at 10mA and using DN2540 at M3, the FET at J1-J4 position could only get a Vds even below 1.5Volts, which can't drive the FETs into saturation area.
If the lower side FET can't go into saturation area, how could it deliver a constant current?
These new SiC JFETs entirely eliminated such problem. The Vgsoff of these transistors is typically -9Volts, which means you'll have over 8Volts for the lower side.
I could say these SiC JFETs fully "liberate" the lower side conponents, providing a better working condition(Over 8.7Volts for 25mA) comparing with traditional Depletion MOSFETs such as DN2540(lower than 1Volts for 25mA) and IXCP10M90S(About 2.4Volts for 25mA). Even LM317 could be used here for a higher current accuracy.
I've bought some samples of the 650V 80mΩ version, and give it to my friend replacing DN2540 in the Mu Follower design. Now his only remainging question is what to use at the lower side😀.
The problems of the SiC JFETs are also clear.
The capacitance is the biggest one. The Rdson=80mΩ version has a Crss of 88pF, while the Crss of Rdson=400mΩ version is 18pF. It's much higher comparing with DN2540.
Maybe packaging the UJ3N1701K2 is the ultimate solution 😵
Here's the curves I traced with DTT-X of UJ3N065080K3S.
These two are the characteristics of UJ3N170400B7S.
And I also managed to trace the output characteristics in high voltage area of UJ3N065080K3S with etracer. Since it's not easy setting, I've only scanned the curves when Vgs=-9.0V.
Notice that Vgs should be provided externally and previously, because the normally-on state of JFETs will trigger the Short Circuit Detection of etracer.
It's not as horizontal as DN2540, so Cascode is necessary.
For example like this, the M3 and J1-J4 formed a Cascode CCS.
If we set it at 10mA and using DN2540 at M3, the FET at J1-J4 position could only get a Vds even below 1.5Volts, which can't drive the FETs into saturation area.
If the lower side FET can't go into saturation area, how could it deliver a constant current?
These new SiC JFETs entirely eliminated such problem. The Vgsoff of these transistors is typically -9Volts, which means you'll have over 8Volts for the lower side.
I could say these SiC JFETs fully "liberate" the lower side conponents, providing a better working condition(Over 8.7Volts for 25mA) comparing with traditional Depletion MOSFETs such as DN2540(lower than 1Volts for 25mA) and IXCP10M90S(About 2.4Volts for 25mA). Even LM317 could be used here for a higher current accuracy.
I've bought some samples of the 650V 80mΩ version, and give it to my friend replacing DN2540 in the Mu Follower design. Now his only remainging question is what to use at the lower side😀.
The problems of the SiC JFETs are also clear.
The capacitance is the biggest one. The Rdson=80mΩ version has a Crss of 88pF, while the Crss of Rdson=400mΩ version is 18pF. It's much higher comparing with DN2540.
Maybe packaging the UJ3N1701K2 is the ultimate solution 😵
Here's the curves I traced with DTT-X of UJ3N065080K3S.
These two are the characteristics of UJ3N170400B7S.
And I also managed to trace the output characteristics in high voltage area of UJ3N065080K3S with etracer. Since it's not easy setting, I've only scanned the curves when Vgs=-9.0V.
Notice that Vgs should be provided externally and previously, because the normally-on state of JFETs will trigger the Short Circuit Detection of etracer.
It's not as horizontal as DN2540, so Cascode is necessary.
Here's what I did with the Semisouth 085 SiC JFET:
https://www.diyaudio.com/community/threads/sic-puppy-se-amp-using-semisouth-085-jfet.192450/
https://www.diyaudio.com/community/threads/sic-puppy-se-amp-using-semisouth-085-jfet.192450/
KoitoYuu ,you can make a good current source with two depletion mosfets, but if you choose them according to Vgs(th) or insert a bias multiplier that will raise the Vds of the lower mosfet.
No need for jfets, SiC or the LM317 which has high noise.
https://www.diyaudio.com/community/threads/tube-i-v-stage.382184/#post-6996594
No need for jfets, SiC or the LM317 which has high noise.
https://www.diyaudio.com/community/threads/tube-i-v-stage.382184/#post-6996594
Bias Multiplyer? You mean using external supplies raising the Vgs like a battery? But that means adding a lot of components here. Just like the Universal Mu Follower I posted. All components on the left side were added just for using the lower side enhancement MOSFET like a Depletion FET. And also such designs will let the gate have a much longer path to ground, together with a higher and more unstable impedance.
I've just tested the LM317 version that I'll use in 300B SE AMP for Constant Current Self Bias.
The current noise of the module is below 100nArms (about 2MHz bandwidth), and even lower than 15nArms if adding 100KHz LP. It'll generate little noise on the cathode capacitor.
Testing conditions: 4ohms shunt resistor, INA849 with 6S1P lithium battery supply and 1000V/V amplification factor. These components formed a 4000V/A current sensing probe.
I have no idea about the noise when using JFETs here, but I think 100nArms is fully acceptable even in a DHT preamplifier.
The bias multiplier does not need a negative voltage, only two additional resistors and a film capacitor .
LM317 has almost 50 times more noise than DN2540, which may not be heard in the cathode, but is unusable in the anode , some other place where low noise is required or high voltage drop , so we can hardly say that LM317 is a universal solution. Two DN2540 in a cascade with a bias multiplier are certainly a better solution than a cascade with LM317. Even one DN2540 is better than a cascade with LM317 .The first time I used the LM317 in cascade in the anode (40x gain) was also the last time because the noise was audible, more than 10 years have passed since then and I have not soldered a single LM317 since then.
https://refsnregs.waltjung.org/ax_High_Perf_Current_Regs_Revisited_0409_052921.pdf
Thanks for your informations. The bias multiplier seems to be an intresting design. I'll consider it in my further applications.
I decided to use LM317 just because I'll use LM317 only at cathode (Current limit of a Shunt Regulator is also acceptable), and only for Constant Current Self Bias. The most severe operating point I will meet is 350mA at about 20-30Volts. The others are typical for 300B SE to 71A SE. It's difficult constructing the lower side without chips like LM317 if I want a high enough DC accuracy at around 350mA. The maximum plate dissipation and grid 2 dissipation of a power tube would be easily exceeded in a class A push pull amplifier.
I won't use LM317 in both Anode loads and Cathode Follower loads. Because when using there, the most significant characteristics are fast response, low capacitance and low noise, while the DC accuracy means nothing. Even a 25% increase in current will not change a lot in most applications, since almost all applications of such designs are used in voltage amplification stages and driving stages. But that will be a total disaster in my application, which is KT170 Class A Parallel Push Pull operating at 435V, 75% UL, 170mA per tube. There'll be two Cascode CCS sharing the total current of 680mA.
Depletion FETs with high enough Idss are rare. Even though there are some, the temperature coefficiency of FETs when operating as a CCS close to Idss is terrible, according to datasheet. And the voltage here is not enough for a Cascode connection of SiC JFETs both entering saturation area.
And by the way, I'm suspecting about the data listed in WaltJung's measuring, since I only measured a current noise of below 15nArms with bandwidth limit from 10Hz to 100KHz, while he said there's more than 1nA/√Hz from 10Hz to 10KHz. Maybe LM317s are much better today? Or, due to the Vgsth of the upper MOSFET, he operated LM317s too close to minimum voltage drop? I'm using the LM317P from ST, whose design is friendly for insulation.
Last edited:
I don't believe that today's LM317 are better than those 10 years ago(today can only be worse).
Today there are much better regulators than LM317, but as far as I know, some of them can be used for precise current sources, but no one uses them in their DIY works.
I looked a bit at what depletion mosfets are available at Mouser and I'm sure that a current source can be made for 1A as well. Maybe even without a bias multiplier if the mosfets are chosen well.
An example is this CCS (without bias multiplier) for 300mA that I made two weeks ago for a regulator in my DAC . The next goal is over 500mA for a new update in the DAC.
Today there are much better regulators than LM317, but as far as I know, some of them can be used for precise current sources, but no one uses them in their DIY works.
I looked a bit at what depletion mosfets are available at Mouser and I'm sure that a current source can be made for 1A as well. Maybe even without a bias multiplier if the mosfets are chosen well.
An example is this CCS (without bias multiplier) for 300mA that I made two weeks ago for a regulator in my DAC . The next goal is over 500mA for a new update in the DAC.
Attachments
Well, It seems that I've used a wrong searching keyword trying to find "Depletion" parts from "MOSFET" section.
There's a lot of them from IXYS. But I mentioned the problem that I've met, the temperature coefficiency.
The Vgs for the same Id of these MOSFETs will increase about 15% as temperature rise from 25℃ to 125℃, which will cause about 10% increase in current.
Maybe I'll try compensating them later, but such compensation only suits a specific current. And there's a lot of works to do.
I'll use Depletion FETs only if the circuit can accept such a poor DC accuracy. For example Anode CCS or Mu Follower of Voltage Amplification stages.
Normally not. It needs a higher voltage across it to regulate well than a Low Dropout Regulator.
TI doesn't spec it like that either.
Jan
TI doesn't spec it like that either.
Jan
Last edited:
LM317 can easily be used as a precise current source, but it depends a bit on how much current you need.
There's some examples in the app section in the datasheet.
In fact, any reg with build-in current limiting can be used as a CCS with suitable resistor.
Jan