Here is my latest design to throw into the ring. It is a 3-stage discrete MC preamp utilizing THAT304 matched PNP-NPN quad as input devices. The remainder of the circuit consists of a cascode VAS stage driving a 4-transistor output stage similar to that used in the OPA627 opamp. It simulates to the same En advertised by THAT Corp: 0.7-0.8 nV/sqrt(Hz). Distortion is in the -120 dB range. There are still a few tweaks to be done (like optimizing input stage resistances to minimize En). I have just about completed PCB layout and should be building the prototype by year's end.
when i open your document i can not see the circuit values very well but i like what i see.
should develop into an interesting project. i ordered samples of the THAT304 but never got them, so i am very interested what you experience.
should develop into an interesting project. i ordered samples of the THAT304 but never got them, so i am very interested what you experience.
Difficulty Reading component Values
Joachim,
The file is a 2003 SP3 Microsoft Word document. When you open it in Word, go to the toolbar and increase the magnification to approx 150% (it is expressed as a percentage). That should fix the problem.
Jeff
Joachim,
The file is a 2003 SP3 Microsoft Word document. When you open it in Word, go to the toolbar and increase the magnification to approx 150% (it is expressed as a percentage). That should fix the problem.
Jeff
Correction to BJT MC Preamp Schematic
I noticed the V+ and V- pin numbers on the IC2 symbody are transposed. V+ should connect to pin 7, and V- to pin 4.
I noticed the V+ and V- pin numbers on the IC2 symbody are transposed. V+ should connect to pin 7, and V- to pin 4.
.pdf Conversion
Thanks for converting the image to a .pdf format. I don't have an Acrobat converter.
Thanks for converting the image to a .pdf format. I don't have an Acrobat converter.
Chassis for Above Described Preamp
After a few days in the shop the chassis for the preamp is nearly complete. Lacking access to sheet metal bending tools I elected to use thick material for the front and side panels and then tap/thread to attach the other panels. I could not find knobs to my liking so I machened the ones in the photo. (It helps to have a vertical mill, a drill press and a lathe). Once all the electronics has been installed and everything tested I'll disassemble the panels and send them to an anodizing shop
After a few days in the shop the chassis for the preamp is nearly complete. Lacking access to sheet metal bending tools I elected to use thick material for the front and side panels and then tap/thread to attach the other panels. I could not find knobs to my liking so I machened the ones in the photo. (It helps to have a vertical mill, a drill press and a lathe). Once all the electronics has been installed and everything tested I'll disassemble the panels and send them to an anodizing shop
Attachments
Could you doublecheck that?
A quick look at the schematic reveals that only the 100ohm resistor in series with the input will add 1.3nV/rtHz. The 30ohm feedback resistor adds another 0.7nV/rtHz and the input stage 3dB more noise than the single ended THAT device (due to the differential configuration).
Total is 1.8nV/rtHz and that's before considering the effect of the emitter degeneration and the current noise effects. Some may find this noise level to high, even for MM cartridges.
The LF gain is set to 60dB, which maps to 40dB @1KHz, after the RIAA correction. This is again good mostly for a MM cartridge (or a high output MC, granted).
I like the case 🙂
Someone incorrectly labled the PDF conversion as a "folded cascode" design.
I can't see any folded cascodes in there at all, but I do see regular cascodes, correctly labled as such by the original designer.
Take note of what Syn says about noise - we don't agree on much but he sure knows his noise calcs!
Regards, Allen
I can't see any folded cascodes in there at all, but I do see regular cascodes, correctly labled as such by the original designer.
Take note of what Syn says about noise - we don't agree on much but he sure knows his noise calcs!
Regards, Allen
Noise Values for Preamp
Please note that the noise voltage described in the first thread was for the input transistors only and was obtained by turning off the noise voltages for the input and feedback resistors. I reran the simulations today with the resistor noise off and got the same value of 0.7 nV/sqrt(Hz).
Also note that the schematic does not list noise optimized values for the feedback and input resistors. When the input and grounded feedback resistors are both set at 50 ohms, the equivalent input noise simulates to approx 1.0 nV/sqrt(Hz) referenced to the input.
The above noise values were obtained by replacing the RIAA network with a simple 10/1 voltage divider, shorting the input resistor to ground, and simulating with a 20 KHz noise BW.
There may be one reason why these numbers seem low. The simulator uses a 1st order LPF to define resistor noise BW, rather than a brickwall filter. I'll try resimulating with a 20 KHz brickwall filter and see if the results change.
Please note that the noise voltage described in the first thread was for the input transistors only and was obtained by turning off the noise voltages for the input and feedback resistors. I reran the simulations today with the resistor noise off and got the same value of 0.7 nV/sqrt(Hz).
Also note that the schematic does not list noise optimized values for the feedback and input resistors. When the input and grounded feedback resistors are both set at 50 ohms, the equivalent input noise simulates to approx 1.0 nV/sqrt(Hz) referenced to the input.
The above noise values were obtained by replacing the RIAA network with a simple 10/1 voltage divider, shorting the input resistor to ground, and simulating with a 20 KHz noise BW.
There may be one reason why these numbers seem low. The simulator uses a 1st order LPF to define resistor noise BW, rather than a brickwall filter. I'll try resimulating with a 20 KHz brickwall filter and see if the results change.
Noise Figures in previous thread
I figured out where the problem may be. If I simulate with a resistor noise BW >> than 20 kHz and then lowpass filter with a 1st order Butterworth characteristic I get the same noise figure as is obtained using the sqrt(4kTBR) expression. So it seems that the simulator's resistor noise definition cuts off far faster than 1st order. In fact it is a comb filter characteristic. For example: using a 100 KHz noise BW and a 20 KHz 1st order Butterworth, a 50 ohm resistor simulates to ~130 nVRMS, while using the above formula gives 127 nVRMS. If a 20 KHz resistor noise BW is specified the simulated noise is ~ 100 nV.
Never trust your sim tools without first trying a few simple examples.
I figured out where the problem may be. If I simulate with a resistor noise BW >> than 20 kHz and then lowpass filter with a 1st order Butterworth characteristic I get the same noise figure as is obtained using the sqrt(4kTBR) expression. So it seems that the simulator's resistor noise definition cuts off far faster than 1st order. In fact it is a comb filter characteristic. For example: using a 100 KHz noise BW and a 20 KHz 1st order Butterworth, a 50 ohm resistor simulates to ~130 nVRMS, while using the above formula gives 127 nVRMS. If a 20 KHz resistor noise BW is specified the simulated noise is ~ 100 nV.
Never trust your sim tools without first trying a few simple examples.
Do not let foul you. Go your own way. THAT is also making transistors to your own design
and if you wish i could help you.
and if you wish i could help you.
Do not let foul you.
Joachim,
No problem, but thanks for the encouragement. I do audio for fun. My day job is running a standards committee for a large American chip company. I'm used to getting pushback.
Joachim,
No problem, but thanks for the encouragement. I do audio for fun. My day job is running a standards committee for a large American chip company. I'm used to getting pushback.
Ok, i am happy that you are so robust. then you do not need my help. i am happy about any well conceaved phonostage and i found your design has potential.
Equivalence of Statistical and Time Averaged Noise
It had occurred to me that the mathematicaly correct definition of noise as a random variable is determined by considering its probability distribution function (PDF) and then computing an RMS value based on +/- one sigma limits of the PDF. I wanted to make sure that this approach yielded the same result as taking the absolute value of the signal and then lowpass filtering it. It turns out that the two methods are equivalent if the noise is truely Gaussian and the PDF is symmetric about zero. See the attached plots as an example. In this example both methods yield approx 1.34 uV RMS noise at the output.
It had occurred to me that the mathematicaly correct definition of noise as a random variable is determined by considering its probability distribution function (PDF) and then computing an RMS value based on +/- one sigma limits of the PDF. I wanted to make sure that this approach yielded the same result as taking the absolute value of the signal and then lowpass filtering it. It turns out that the two methods are equivalent if the noise is truely Gaussian and the PDF is symmetric about zero. See the attached plots as an example. In this example both methods yield approx 1.34 uV RMS noise at the output.
Attachments
PCB back, Stuffed, and Under Test
Over the last two weekends I got the PC boards back. they consist of one PCB for a remote power supply plus a PCB for the phono and lineamp. The latter two utilize similar layouts, but with different component values - particularly in the feedback loop.
The power supply is straightforward, consisting of LM317/LM337 regulators with ~3000 uf filter caps on the output. I used a C-L-C network from the bridge rectifier, because it removes approx 20x more of the 120 cycle noise than a single stage cap filter.
The lineamp circuits powered up without difficulty. the only change required was to lower the dominant pole slightly to fix an HF oscillation that occurred at the onset of clipping. Even so, tR and tF are still ~300 ns for a 10 VPP swing into a 600 ohm load. With the input shorted the ouput noise was near the limit of my Keithley 5 1/2 digit voltmeter, <10 uVRMS. The phono stages powered up without any stability problems at all. The next step is to check the accuracy of the RIAA network and then put the PCBs into the chassis I built.
There is still one more PCB to be fabricated. On it are directly mounted the RCA jacks and the dry reed relays used for signal switching. This approach allows for the use of low cost panel-mounted switches and saves in wiring complexity. It should also improve signal integrity.
Over the last two weekends I got the PC boards back. they consist of one PCB for a remote power supply plus a PCB for the phono and lineamp. The latter two utilize similar layouts, but with different component values - particularly in the feedback loop.
The power supply is straightforward, consisting of LM317/LM337 regulators with ~3000 uf filter caps on the output. I used a C-L-C network from the bridge rectifier, because it removes approx 20x more of the 120 cycle noise than a single stage cap filter.
The lineamp circuits powered up without difficulty. the only change required was to lower the dominant pole slightly to fix an HF oscillation that occurred at the onset of clipping. Even so, tR and tF are still ~300 ns for a 10 VPP swing into a 600 ohm load. With the input shorted the ouput noise was near the limit of my Keithley 5 1/2 digit voltmeter, <10 uVRMS. The phono stages powered up without any stability problems at all. The next step is to check the accuracy of the RIAA network and then put the PCBs into the chassis I built.
There is still one more PCB to be fabricated. On it are directly mounted the RCA jacks and the dry reed relays used for signal switching. This approach allows for the use of low cost panel-mounted switches and saves in wiring complexity. It should also improve signal integrity.
- Status
- Not open for further replies.