Announcement

Collapse
No announcement yet.

Why some amps use two triode in parallel?

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Oh yeh, what we are talking is only a narrow facet on the parallel tube. I did say it sounds better. Only disagreement is about the noise.

    Comment


    • its also a good way to use more tubes since some people count the tubes and pay accordingly...Fryette's Deliverance uses this too.

      Comment


      • I set up an experiment as follows. A small chasses was prepared with two 9 pin ceramic sockets. One socket was wired with both triodes in parallel, the other just one triode was wired. The single socket had a 1.5K CF cathode resistor bypassed with 100uF. The parallel socket had a trim pot has the cathode resistor bypassed with 100uF cap. The single socket's plate voltage was 208V with the 12AX7LPS so the trim pot was adjusted for the same 208V when the 12AX7LPS was installed in the parallel socket. Both sockets were wired to a single 100K 1W CF plate resistor. Both sockets were wired to the same 33K CF grid resistor connected to ground. To change from single to parallel triodes only required moving the tube being tested to the other socket. The same grid and plate resistors were used in either case.

        When measuring gain the generator drives the 33K just fine, when measuring noise the generator was programmed to zero volts so the grids were effectively grounded by the generator's 600 ohm output impedance. The generator was disconnected to measure the noise with the 33K resistor.

        The output of the tube was measured by connecting a 10Meg scope probe to the input of a Tektronix 7A22 plugin. The plugin was set to 30KHz bandwidth, the low frequency limit set by the AC coupling cap is less than 1 Hz. The measured gains might seem high but that is because of the 10Meg load. The scope's vertical output was connected to an Agilent 3458A meter to get a true RMS value of the noise. A spreadsheet was used to correct for the scope's gain depending on the input sensitivity seting and the scope output gain. Noise measurements were made on the 500uV per division range.

        The spreadsheet divided the measured noise by the measured gain to get the equivalent input noise. The heater to the tubes was powered by a regulated DC bench supply set to 6.30V. The B+ of 300V came from a regulated bench supply and further decoupled by a 10H choke and 10uF cap. Noise outputs were on the order of 1 to 2 mV peak to peak at the plate. Getting all the line noise out of the output was a challenge. The scope had to be connected via differential probes to get rid of several milliVolts of ground noise. Grounding the table everything was setting on really helped.
         
        Sovtek 12AX7 LPS

        Single triode
        Gain: 61.4
        Equivalent input noise input shorted: 2.67uV
        Equivalent input noise input 33K: 4.4uV

        Paralllel triodes
        Gain: 69.5
        Equivalent input noise input shorted: 2.12uV (-2.02dB)
        Equivalent input noise input 33K: 4.12uV (-0.6dB)
         
        Made in China 12AX7A

        Single triode
        Gain: 63.3
        Equivalent input noise input shorted: 2.34uV
        Equivalent input noise input 33K: 4.70uV

        Paralllel triodes
        Gain: 66.9
        Equivalent input noise input shorted: 2.67uV (+1.14dB)
        Equivalent input noise input 33K: 4.58uV (-0.23dB)
         
        Slightly used Telefunken 12AX7A

        Single triode
        Gain: 58.1
        Equivalent input noise input shorted: 2.71uV
        Equivalent input noise input 33K: 4.33uV

        Paralllel triodes
        Gain: 61.1
        Equivalent input noise input shorted: 2.30uV (-1.42dB)
        Equivalent input noise input 33K: 4.34uV (+0.02dB)
         
        NOS GE 12AX7A/7025

        Single triode
        Gain: 61.4
        Equivalent input noise input shorted: 2.96uV
        Equivalent input noise input 33K: 5.44uV

        Paralllel triodes
        Gain: 66.5
        Equivalent input noise input shorted: 2.51uV (-1.43dB)
        Equivalent input noise input 33K: 5.26uV (-0.3dB)

        The measured numbers indicate that input noise is dominated by the grid stopper and only a small improvement in signal to noise ratio is obtained by parallel triodes unless the grid stopper is eliminated.
        WARNING! Musical Instrument amplifiers contain lethal voltages and can retain them even when unplugged. Refer service to qualified personnel.
        REMEMBER: Everybody knows that smokin' ain't allowed in school !

        Comment


        • Originally posted by loudthud View Post

          Sovtek 12AX7 LPS

          Single triode
          Gain: 61.4
          Equivalent input noise input shorted: 2.67uV
          Equivalent input noise input 33K: 4.4uV

          Paralllel triodes
          Gain: 69.5
          Equivalent input noise input shorted: 2.12uV (-2.02dB)
          Equivalent input noise input 33K: 4.12uV (-0.6dB)
           
          Made in China 12AX7A

          Single triode
          Gain: 63.3
          Equivalent input noise input shorted: 2.34uV
          Equivalent input noise input 33K: 4.70uV

          Paralllel triodes
          Gain: 66.9
          Equivalent input noise input shorted: 2.67uV (+1.14dB)
          Equivalent input noise input 33K: 4.58uV (-0.23dB)
           
          Slightly used Telefunken 12AX7A

          Single triode
          Gain: 58.1
          Equivalent input noise input shorted: 2.71uV
          Equivalent input noise input 33K: 4.33uV

          Paralllel triodes
          Gain: 61.1
          Equivalent input noise input shorted: 2.30uV (-1.42dB)
          Equivalent input noise input 33K: 4.34uV (+0.02dB)
           
          NOS GE 12AX7A/7025

          Single triode
          Gain: 61.4
          Equivalent input noise input shorted: 2.96uV
          Equivalent input noise input 33K: 5.44uV

          Paralllel triodes
          Gain: 66.5
          Equivalent input noise input shorted: 2.51uV (-1.43dB)
          Equivalent input noise input 33K: 5.26uV (-0.3dB)

          The measured numbers indicate that input noise is dominated by the grid stopper and only a small improvement in signal to noise ratio is obtained by parallel triodes unless the grid stopper is eliminated.
          One note, even if you set the generator to 0V, there is still noise from the generator and also the miller caps at the input forms a low pass filter with the impedance of the generator that limit the BW of the parallel triodes more. You should literally short out the input to be fair.

          Also, I think you should adjust the pot at the cathode of the parallel so the current of the single and parallel is the same. This is important as the current produces shot noise and if you have different current, you are not having the same condition.

          But let's look at the result of the shorted input: 3 out of 4 indicates between 1 to 2 db lower using parallel triodes. But the Chinese is actually higher with the parallel!!! If the measurement is correct, it must be the flicker noise contribution that is not gain independent.

          From my calculations in post #47, if you remove the grid stop and short the grid to ground, the noise due to thermal and current shot noise should be lower than your measurement. So your measurement must have a lot of the flicker noise in it.



          It still does not make sense with your result. The average gain only increase from about 60 to 67 going from single to parallel. Which is about 1dB gain. According to theory, the best case is noise increase by 3dB and gain increase by 6dB. Theory concludes you have 3dB gain in signal to noise ratio from that. BUT you only shows about 1dB voltage gain, not 6dB. I don't understand why you get lower noise with parallel.

          I really think you should redo the measurement with real shorted input with no grid stop resistor so you really have 0 ohm from grid to ground. I think your generator output might have something to do with it as it does not make sense. I suspect the input impedance of the generator and the miller cap forms a low pass. The miller cap is bigger for parallel triodes and limited the bandwidth to half. Since you set the noise BW to 30KHz, the added miller cap can lower the bandwidth enough to give you better result in the parallel case. Noise calculation is tedious.


          EDIT: I edited the post after thinking about it for a few hours.
          Last edited by Alan0354; 03-24-2014, 10:20 AM.

          Comment


          • Originally posted by Alan0354 View Post
            I really think you should redo the measurement with real shorted input with no grid stop resistor so you really have 0 ohm from grid to ground.
            Agreed. An EIN of >2uV is unusually high for any triode (I predict a little under 1uV), and the gain of the parallel triodes is unusually low. Something's going on here...

            Comment


            • The residule noise of my setup was about 1uV refered to input. Shorting the input to ground didn't change the noise by much. I could have used the AP System 1 to measure the noise but it has an input impedance of about 100K. Not really suitable for a vacuum tube circuit. Then there is the issue of blowing up my equipment. Thes things really don't like big voltage transients. I tried just strapping two triodes with the typical 100K/1.5K setup in parallel and the gain only increased by 0.1dB. I think there was a noise reduction in my setup due to the reduced current each triode. If you think you can increase the gain by 6dB, show me the circuit after you have prototyped it.
              WARNING! Musical Instrument amplifiers contain lethal voltages and can retain them even when unplugged. Refer service to qualified personnel.
              REMEMBER: Everybody knows that smokin' ain't allowed in school !

              Comment


              • The note about how they measure noise on the RCA 7025 data sheet doesn't make any sense to me. They specify a plate load resistor of 2700 ohms ??? I think they meant 2700 ohm cathode resistor, but then what was the plate resistor?
                WARNING! Musical Instrument amplifiers contain lethal voltages and can retain them even when unplugged. Refer service to qualified personnel.
                REMEMBER: Everybody knows that smokin' ain't allowed in school !

                Comment


                • The 7025 RCA datasheet says the plate is 0.1M, and the cathode is 2.7k. In any case, see if you can duplicate the results for the 12AX7, from the GE datasheet: Rp=100k, Rk=2.7k, Ck = 100uF, Rg = 0, BW = 25-10kHz.

                  Comment


                  • Originally posted by loudthud View Post
                    The residule noise of my setup was about 1uV refered to input. Shorting the input to ground didn't change the noise by much. I could have used the AP System 1 to measure the noise but it has an input impedance of about 100K. Not really suitable for a vacuum tube circuit. Then there is the issue of blowing up my equipment. Thes things really don't like big voltage transients. I tried just strapping two triodes with the typical 100K/1.5K setup in parallel and the gain only increased by 0.1dB. I think there was a noise reduction in my setup due to the reduced current each triode. If you think you can increase the gain by 6dB, show me the circuit after you have prototyped it.
                    If you measure the same result with shorted input, there must be other reason why. I am glad Mr. Merlinb joined in. My calculation ONLY include thermal and shot noise. There is a big factor of flicker noise I have no idea about the value. AND to the best of my knowledge, the grid input leakage current is 1uA and it does not change with operating points.

                    A) So the questions to Mr. Merlinb are:

                    1) What is the input grid current characteristics when in normal small signal where grid is more negative than cathode.

                    2) What is the flicker noise model? If not model, what is the characteristics?

                    3) What other sources of noise other than what I described?



                    B) Also you need to measure the -3dB frequency response of the DUT in both cases. Noise has to be defined with bandwidth. In parallel case, Miller cap is DOUBLED. So this can act as a low pass filter and lower the BW by half. This will automatically show false advantage to the parallel triodes. Only way to test is lower the BW of your instrument to 7-10 KHz as the amp will not produce sound beyond 6 or 7 KHz.

                    C) As for protection of your instrument, When I do noise measurement, I usually use a gain block of at least 40dB ( 100) following the DUT ( device under test) to amplify the noise before driving into any instrument. I never have fancy instruments to measure noise. With that, you literally look at it on the scope!!! I use opamp in non inverted configuration to get input impedance way in Mohm to buffer the DUT and amplify the noise signal. Using opamp protect your instrument from high voltage also. Your instrument will never see the HV.

                    Noise measurement is very tedious, one need to go over and over the setup to make sure everything is correct. Been doing this for long time and from my experience the calculation work if you have the complete noise model of the DUT and you can reproduce with measuring equipment. If you do get the result from the measurement, you are missing something or the model is wrong. You just need to know all the conditions.

                    Comment


                    • Originally posted by Alan0354 View Post
                      A) So the questions to Mr. Merlinb are:

                      1) What is the input grid current characteristics when in normal small signal where grid is more negative than cathode.
                      Grid current is normally between 1 and 100nA, and grid current shot noise can be negelected for source impedances less than about 10k.

                      2) What is the flicker noise model? If not model, what is the characteristics?
                      Input referred flicker noise is approximately described by:
                      EIN = { [K Ia^2 ln(fhi/flo) ]^0.5 } / gm

                      Where:
                      K is about 10^-13.
                      Ia = anode current in amps
                      fhi = upper bandwidth limit
                      flo = lower bandwidth limit

                      This gives a corner frequency smack in the middle of the audio band, usually.

                      3) What other sources of noise other than what I described?
                      Nothing you haven't mentioned. The principle sources are:
                      Anode current shot noise, EIN = {(2.8E-20 * gm * B)^0.5 } / gm
                      Anode current flicker noise, EIN see above
                      Grid current shot noise = (2 * q * Ig * B)^0.5
                      There is of course Johnson and excess noise in the anode resistor, but it will be negligible compared with valve noise (except for carbon composition).


                      If you think you can increase the gain by 6dB, show me the circuit after you have prototyped it.
                      Yeah, but you're getting absurdly little increase. Typically the gain jumps from about 60 to about 70 when going from single to parallel 12AX7s.

                      The EIN voltage of any triode is seldom over 1uV in the audio band, which would amount to about 60uV at the anode in your case. I doubt you can measure that reliably in a 500uV/div range.
                      Last edited by Merlinb; 03-25-2014, 10:04 AM.

                      Comment


                      • Originally posted by Merlinb View Post
                        Grid current is normally between 1 and 100nA, and grid current shot noise can be negelected for source impedances less than about 10k.
                        Input referred flicker noise is approximately described by:
                        EIN = { [K Ia^2 ln(fhi/flo) ]^0.5 } / gm
                        Where K is about 10^-13.
                        This gives a corner frequency smack in the middle of the audio band, usually.
                        EIN = { [K Ia^2 ln(fhi/flo) ]^0.5 } / gm

                        1) What is "a", fhi and flo?

                        2) Is "I" the grid current that is normally between 1nA and 100nA?

                        3) If EIN is the input referred noise, where is the referred flicker noise from the anode current? Or because of the gain is greater than 60, you ignore the contribution of the anode current?

                        Thanks

                        Comment


                        • Ia is anode current, a = anode.
                          fhi, flo = upper and lower frequency, i.e., the high and the low frequency of interest.respectively

                          Comment


                          • Originally posted by Alan0354 View Post
                            EIN = { [K Ia^2 ln(fhi/flo) ]^0.5 } / gm
                            1) What is "a", fhi and flo?
                            Sorry. I have editied my earlier post. As Alan said,
                            Ia = anode current in amps
                            fhi = upper bandwidth limit
                            flo = lower bandwidth limit

                            Comment


                            • Top effort Gordon. Did you find that the measurements were repeatable within say +/- 0.1uV of EIN ? Were you physically holding the probe during measurements? Was the heater supply floating at the supply end, but grounded at the 0V test circuit end? Perhaps the only easy way to exclude one parasitic grounding loop would be to use a 12V battery and linear reg for the 6.3V heater.

                              Comment


                              • It is hard to speculate where the noise come from. So far we are calculate noise JUST from the circuit, we did not include any radiated noise or ground power noise.

                                That's the reason I suggested using a gain block. You cannot see anything using a scope at the output of just one stage. With a X100 opamp gain block, you can use a BNC direct into the scope and you can see right away whether you have any 60, 120 etc. noise from the power. Then fix it accordingly. In my case, I don't have the right setup. I only look at the high gain channel and I use rectified DC ( not regulated) to quiet the power noise down enough. But it's still not there.

                                Using a scope is surprising good looking at noise if you have gain block. You troubleshoot the unwanted noise before using a noise meter. Noise meter does not differentiate power noise from tube characteristic noise that we are talking about.

                                So far, both me and Merlinb feel the measured noise of the shorted input is WAY WAY too high to be right. Use gain block and fix the other noise first. Or else, we could be theorizing to death on the noise that has nothing to do with the noise characteristics between single and parallel triodes.

                                Comment

                                Working...
                                X