Introduction
This is a continuation of my
previous blog post. I will try to share my experience through the design process of this RIAA stage through these individual posts with an attempt to spark some interest in others and in return to get some valuable input from the experience and knowledge of others. Hope this works!
First stage
Here is the initial design version for analysis. I’m working through this step by step and refining the circuit in every iteration. The initial circuit is very simple. The first stage is key. We want to achieve as much amplification as possible from this stage before we hit the LCR network. The choice of the
6S17K-V valve may appear as a surprise to the ones not familiarised with this valve. Here are some notes from Wavebourn around this valve:
“[…]the tube has an amplification factor of 135 (higher than 12AX7),
transconductance of 14 milliamperes / volt (8 times more than 12AX7);
that means higher amplification factor with lower noises and output resistance.
And lower capacitances, because the tube was made to work on frequencies up to microwaves. […]
It’s planar frame grid is made of microscopic wires 8 microns thin, with distance between wires 18 micron.
Cathode emission is strong, and the grid is close to the cathode, so grid current is very high compared to 12AX7!
100 KΩ grid leak resistor would cause the tube to cutoff!
Also, cathode is connected to one end of filament. That means, it needs a different bias scheme.
Cathode should be completely grounded, no cathode bias resistor is needed.
It is good because it eliminates hum caused by leakage of filament voltage on cathode.
Also, it eliminates the need to shunt cathode bias resistor by a capacitor that can’t be ideal.
How to bias the tube if it’s cathode is grounded?
It can bias itself due to input current that causes voltage drop
in secondary winding of an input transformer and a grid stopper resistor.
Grid stopper resistor can be used to adjust bias hence idle current. This resistor can be in kΩ value, that means very low added by it noises.
Compare to 10MΩ needed to self-bias a venerable EF86! […]”
I traced this valve some years ago and produced an initial Spice model. A 40dB stage gain can be achieved with this planar triode easily. In my circuit, Rg is the grid stopper resistor which provides the bias to the valve. R1 should be the recommended load resistance for the cartridge. In my case should be a 200Ω resistor or more. Since this initial design doesn’t provide enough gain for the DL103, an input step-up transformer is missing. In which case, R1 would be the required secondary load resistance suitable for the transformer. I will come back to this point later on (probably in a different post). The gain of this first stage is established by the anode resistance (R2), the anode resistance (Ra) and the load resistance (R9) in addition to the mu factor. We want to use a low noise resistor (e.g. wire-wound) for R2. The LF response will be dominated by the pole formed by C6 and R9 in addition to the output impedance of the valve, so we want to keep R9 as high as possible. There is a limitation in terms of the acceptable grid leakage resistor value. The higher the value of R9, the lower the value of C6 (the smaller the cap the better it is), gain is maximised and distortion is minimised. Surely the initial values used here can be optimised. We will come back to this later.
In the above plot we can see that at 57V anode voltage and 5.3mA anode current the gain of this triode is approximately 192 (Note that this is higher than the μ reported by Wavebourn above but looking at the datasheet we can see a significant variance in this parameter so we should expect a real challenge in matching a pair of these tubes at least). Transconductance is high at 14mA/V and anode resistance is 13KΩ. The gain of this stage is therefore:
Av=μ * RL / (RL+Ra) whereas RL = 27KΩ ||100KΩ = 21,260Ω so Av=118.85 or 41.5dB.
Ideally with an CCS load we could achieve at least a gain of 170 (44.6dB). However, I suspect the noise level with the CCS would be a real challenge in practice.
The cathode follower
The cathode follower is formed by the famous 6C45P-E. A tricky fellow which has a reputation to oscillate widely given it’s characteristics. This chap has a gm of about 45mA/V which will provide an output impedance of 1/gm in parallel with the cathode resistor. About 20Ω in our case. So this means the additional resistor needed to match the LCR impedance has to take into account this intrinsic output impedance of the cathode follower which is the driver of the LCR network. The value of R8 is actually 6KΩ minus the 20Ω of the follower.
An interesting point to add here is that although a 600Ω load is not impossible for a 6C45P-E cathode follower, it’s definitely a heavy one. And this will be reflected in the distortion. I’ve done some simulations to compare the distortion impact of an 6KΩ versus 600Ω LCR and it seems on paper that there is a penalty of 0.01% in distortion. Not trivial so it’s worth looking at an LCR with 6KΩ or more.
The final amplification stage
The final stage needs to provide the additional gain required considering the 20dB insertion loss of the LCR network. The additional challenge for this stage is to achieve very low distortion at the desired output levels. I decided to use the
6J52P which is similar to the D3a simply because I have many of them, they do sound good and also traced the curves in triode mode not far ago:
With an CCS load, a gain closely to mu can be achieved subject to the output RL. Using a pair of SiC diodes for bias and a low anode current of about 15mA, the gain of the stage with 100KΩ load will be around 34.5dB. The anode resistance of the 6J52P is about 2KΩ so we need a bigger output capacitor which isn’t great. We can improve this (and we will do this later).
A simple coupling option for the LCR into the 6J52p is a capacitor. Unfortunately again, this capacitor has to be of a high value considering the 6KΩ LCR network and the grid leakage resistor (RL in the circuit above). There are several option to consider, like a grid choke to reduce the load impedance to the LCR network or DC coupling. Both option will be explored later.
I ran out of time so will leave more than 10 iterations of this design for later…
Related
Ale I have a newbie question. Given that the 6C45P-E cathode follower Zout is 20 ohms and a 600 ohm Zin is 30 times this, why do you characterize this as a heavy load? Is it because of the idle current being “low”? What makes a load heavy or light?
Nice, but I wonder why not remove C5 (its big and bulky) and add three more diodes to the cathode of U3 (Di, d2 + D3, D4 and D5)?
Input capacitance is not that big and might work for MC via a SUT…
Yes, you can remove C5 indeed and adding additional bias to the 6Z52P is ok at low signal levels. In fact the anode CCS will force the operating point and it’s AC-coupled at the output.
I remember when I rigged this up, the cathode follower was replaced with a source follower and it was DC-coupled in the end as the bias variance was adjusted with a SiC array in the cathode and adjusting the anode CCS current as you suggested. It worked well both ways and I couldn’t notice a big sound difference. As you say, best to remove the cap if possible.