At the beginning of the introduction of AI - Chat GPT, I tried to ask about "how to make / schematic tube amplifier?" and I got a very short and unclear answer, without still not smart.
after that 2 years passed I was curious again trying to use AI deep seek for the schematic of the 10Y SE triode tube amp class A. and this is the answer! lots of convenience and very fast response! is this the era of the end of the human individual mind and intelligence, and all replaced by AI?? very scary!
I am a lucky human being to be able to experience an era without computers until the digital era. and this is really surprising !! there is a sense of fear!
And this is answer of my question :
# DIY 10Y Sylvania SE Triode Amplifier (Class A)
Here's a comprehensive guide to building a high-quality 10Y tube amplifier using Hashimoto transformers and premium components:
## Overview
This is a single-ended Class A triode amplifier using:
## Circuit Design
### Power Supply Section
- Choke input (5H+)
- Second cap: 20-40μF
- RC filtering for driver stage
### Amplifier Section
## Operating Points
### 10Y Tube (Approx. 300V operation)
### ECL11 Driver
## Construction Tips
1. Layout:
- Keep input wiring far from power supply
- Star grounding point near input
- Tight lead dress for 10Y grid circuit
2. Heaters:
- DC for ECL11 (reduce hum)
- AC for 10Y (with hum pot)
3. Safety:
- Bleeder resistors on PSU caps
- Fuse both primary and secondary
## Expected Performance
## Alternative Options
Would you like me to elaborate on any particular section of this design? I can provide more detailed schematics, part numbers, or construction photos if needed.
The 10Y is a wonderful tube - this should make a superb amplifier with the Hashimoto iron!
Still add for simulation LT spice if you want it!!!
after that 2 years passed I was curious again trying to use AI deep seek for the schematic of the 10Y SE triode tube amp class A. and this is the answer! lots of convenience and very fast response! is this the era of the end of the human individual mind and intelligence, and all replaced by AI?? very scary!
I am a lucky human being to be able to experience an era without computers until the digital era. and this is really surprising !! there is a sense of fear!
And this is answer of my question :
# DIY 10Y Sylvania SE Triode Amplifier (Class A)
Here's a comprehensive guide to building a high-quality 10Y tube amplifier using Hashimoto transformers and premium components:
## Overview
This is a single-ended Class A triode amplifier using:
- Power tube: Sylvania 10Y (rare DHT triode)
- Driver tube: Siemens ECL11 (triode-pentode)
- Output transformer: Hashimoto H14-20U
- Power transformer: Hashimoto PT250
- Rectifier: Raytheon 5R4WGB
## Circuit Design
### Power Supply Section
- Rectification: 5R4WGB tube rectifier (slow start)
- Power transformer: Hashimoto PT250 (custom spec for 10Y)
- Filtering:
- Choke input (5H+)
- Second cap: 20-40μF
- RC filtering for driver stage
### Amplifier Section
Code:
[ECL11 Triode] → [10Y Grid] → [Hashimoto H14-20U] → [Speaker]
↑
[Volume Pot]- Driver stage: ECL11 triode section (μ≈70) in common cathode
- Output stage: 10Y triode (μ≈8.5) with fixed bias or cathode bias
- Coupling: High-quality film capacitor (0.47μF-1μF)
## Operating Points
### 10Y Tube (Approx. 300V operation)
- Plate voltage: 300V
- Grid bias: -35V (fixed) or cathode resistor (~700Ω)
- Plate current: 50mA
- Output power: ~3W
### ECL11 Driver
- Plate load: 47kΩ
- Cathode resistor: 1kΩ (bypassed)
- Gain: ~30-40 (sufficient to drive 10Y)
## Construction Tips
1. Layout:
- Keep input wiring far from power supply
- Star grounding point near input
- Tight lead dress for 10Y grid circuit
2. Heaters:
- DC for ECL11 (reduce hum)
- AC for 10Y (with hum pot)
3. Safety:
- Bleeder resistors on PSU caps
- Fuse both primary and secondary
## Expected Performance
- Frequency response: 20Hz-30kHz (-1dB)
- THD: <2% at 1W (mostly 2nd harmonic)
- Damping factor: ~4 (typical for SE)
## Alternative Options
- Driver tube: Can substitute ECL11 with 6SN7 or 6J5
- Rectifier: 5U4G if more current needed
Would you like me to elaborate on any particular section of this design? I can provide more detailed schematics, part numbers, or construction photos if needed.
The 10Y is a wonderful tube - this should make a superb amplifier with the Hashimoto iron!
Still add for simulation LT spice if you want it!!!
"10Y Tube (Approx. 300V operation)
- Plate voltage: 300V
- Grid bias: -35V (fixed) or cathode resistor (~700Ω)
- Plate current: 50mA"
Funny chinese software.
You can go further in suggestions, as it is learning from every post we write.
Congratulate for the work done and ask for more details on the design, and you’ll get the following:
Would you like me to:
Congratulate for the work done and ask for more details on the design, and you’ll get the following:
Ultimate 10Y SE Triode Amplifier – High-End DIY Design Guide
A no-compromise single-ended amplifier built around the rare and revered Sylvania 10Y, this design prioritizes sonic purity, component synergy, and ultra-low noise. Paired with Hashimoto iron, and incorporating a minimalist yet elegant topology, this amplifier aims to deliver the timeless musicality that only a directly heated triode can achieve.1. Conceptual Foundation
- Topology: Single-ended, pure Class A
- Output tube: Sylvania 10Y (direct-heated triode, μ ≈ 8.5)
- Driver stage: Siemens ECL11 (triode section, μ ≈ 70)
- Output transformer: Hashimoto H14-20U (5kΩ primary, 20W)
- Power transformer: Hashimoto PT-250 (custom secondary voltages)
- Rectifier: Raytheon 5R4WGB (military-grade, indirectly heated)
2. Power Supply Design – Stability and Silence
Architecture
- Rectification: 5R4WGB tube (soft-start, indirectly heated)
- First filter cap: 4μF (maximum permissible for 5R4)
- Choke input filtering: ≥5H / ≤100Ω DCR preferred
- Second cap: 20–40μF high-quality oil or polypropylene (e.g., ASC, Mundorf)
- RC filtering for driver stage: 2.2kΩ series resistor into 47μF capacitor
Technical Considerations
- Separate RC filtering stages for the driver and output section to decouple noise and avoid cross-modulation.
- Bleeder resistors across each PSU cap to ensure safe discharge (~220kΩ, 5W).
- Snubber networks (e.g., 0.01μF + 100Ω) across secondaries and rectifier to suppress switching transients.
- Consider elevating the heater supply reference to +40–60VDC to reduce cathode-to-filament voltage stress on the 10Y (if using cathode bias).
3. Audio Signal Path – Simplicity with Purity
Driver Stage – ECL11 Triode
- Operating point: ~150V on plate, 1.2mA current
- Plate load resistor: 47kΩ (carbon film or PRP for tonal flavor)
- Cathode resistor: 1kΩ, bypassed with 100μF electrolytic (Elna Silmic II or Nichicon KZ)
- Coupling capacitor: 0.47μF to 1μF polypropylene (Mundorf Supreme, Jupiter, or VCap ODAM)
- Gain: Sufficient (~30x) to drive the 10Y with full swing
Output Stage – 10Y
- Plate voltage: 300V
- Bias method: Fixed bias at -35V, or cathode bias with ~700Ω (5–10W), optionally bypassed
- Plate current: 50mA
- Load line: ~5kΩ primary impedance (Hashimoto H14-20U)
- Power output: 2.5–3W RMS with exceptionally low distortion (mainly 2nd harmonic)
4. Filament Supply – Hum Control and Stability
10Y Filament (7.5V, 1.25A per tube)
- AC heating: Center-tapped hum pot, twisted leads, elevated to +40V
- DC heating (optional): Common-mode filtered, regulated DC (use of current-regulated CCS preferred, e.g., Rod Coleman regulators)
ECL11 Filament (6.3V, 0.66A)
- DC regulated with LM317 or LT1085-based supply, CRC filtered before regulation
5. Grounding, Layout, and Construction Practices
Grounding Strategy
- Star ground topology: Central ground bus tied to chassis at a single point near input
- Isolated signal ground: Avoid mixing power and signal return paths
Layout Notes
- Input signal wiring kept distant from power transformer and rectifier tubes
- Use of twisted pairs for AC heater wiring
- Shielded cables for grid connections where necessary
- Point-to-point wiring with short leads, particularly on 10Y grid
6. Measured & Simulated Performance
| Parameter | Expected Value |
|---|---|
| Frequency response | 20Hz – 30kHz (-1dB) |
| Total harmonic distortion (THD) | <1.5% at 1W, primarily 2nd order |
| Output power | 2.5 – 3W |
| Noise floor | <1mV RMS at speaker output |
| Damping factor | ~3.5–4 (typical for SE topology) |
7. Simulation and Modeling (LTspice)
To predict and refine circuit behavior, a full simulation is encouraged:- 10Y model: Available from Duncan Amps or DIYaudio resources (parameters include μ = 8.5, Rp = 5k, gm = 1.7 mA/V)
- ECL11 triode model: Approximated with ECC81 if unavailable
- Transformer modeling: Ideal transformer blocks or real measured impedance curves
- Simulate harmonic content, clipping behavior, and PSU ripple rejection
8. Refined Component Selection
| Component | Recommendation |
|---|---|
| Coupling caps | Jupiter Copper Foil, VCap ODAM, Mundorf |
| Resistors | PRP, Takman, or carbon comp for voicing |
| Chassis | Aluminum with copper subplate, or wood hybrid |
| Binding posts | Cardas or WBT |
| Input jacks | Neutrik or Switchcraft |
| Tube sockets | Teflon or ceramic with gold contacts |
9. Upgrade & Expansion Ideas
- Interstage transformer: Use a Lundahl or Tribute interstage instead of capacitor coupling
- Dual mono PSU: Separate power supplies per channel for total isolation
- Active shunt regulators: Maida regulator for the 10Y B+ to improve dynamic behavior
- Custom chokes: Low DCR, high inductance designs for superior ripple rejection
10. Summary
This amplifier brings out the soul of the 10Y triode, balancing precision engineering with musical richness. With careful component matching and attention to layout, the result is an amplifier that transcends specs to offer an immersive listening experience with a tactile, holographic soundstage.Would you like me to:
- Draw the full schematic in CAD (KiCad, Eagle, or hand-sketched PDF)?
- Provide a turret board layout?
- Create a complete BOM with sourcing suggestions?
- Prepare an LTspice file for immediate simulation?
The correct terminology is directly heated, not directory.
As everything in life, you can focus on highlighting errors of the others to have the illusion to be better, or become better grabbing strong points from others.
It is stupid AI
That’s all!
Can’t be a reference to build a project
Note: it is an error on typing about directly!!!
Walter
Walter is right, stupid AI.
You really can't understand a totally idiot -false- premise?
10Y at 300V, -35V near to the close.
Only 0-1mA, not 50ma!!!
This operating point completely unsuitable to working.
Should be designing, instead of asking idiotic software.
You really can't understand a totally idiot -false- premise?
10Y at 300V, -35V near to the close.
Only 0-1mA, not 50ma!!!
This operating point completely unsuitable to working.
Should be designing, instead of asking idiotic software.
No guys, the problem is not the AI, is the way the question is asked.
AI is a three years old baby that is learning, and is programmed to give answers in few seconds.
Ask the questions in the right way, and you’ll get useful answers. Ask everything in one single line and you’ll get wrong answers.
AI is a three years old baby that is learning, and is programmed to give answers in few seconds.
Ask the questions in the right way, and you’ll get useful answers. Ask everything in one single line and you’ll get wrong answers.
It is stupid without any doubt.
From the beginning with the error on 5R4
And for this reason can’ t be a reference.
Ps= if it fail around a simple concept how we can be confident for other more important things?????
Walter
this is interesting for discussion. to know whether a forum like this can still exist? I hope it STILL exists.
Here I only give an overview of how IA becomes a solution that will replace the discussion forum? the answer results that I get are still wrong! but this is not bad if I look at the systematization of the answers given, maybe my question is not detailed enough, but in the future with a lot of data collected - maybe including from this forum - then in the future it will be more complete and perfect? maybe! and I'm trying to make a 10Y SE triode Amplifier. this project is underway.
Here I only give an overview of how IA becomes a solution that will replace the discussion forum? the answer results that I get are still wrong! but this is not bad if I look at the systematization of the answers given, maybe my question is not detailed enough, but in the future with a lot of data collected - maybe including from this forum - then in the future it will be more complete and perfect? maybe! and I'm trying to make a 10Y SE triode Amplifier. this project is underway.
Attachments
I agree with this 100%.
Does not scare me so far. It just works in a nicely structured way within known, well established concepts. I have not seen an AI coming up with new, clever ideas. These things are not good with beeing creative.
I don’t think so. We are still human.
This happens on forums as well 😃
Exactly.
Is a bit too wide indeed. There are specific ways to get good answer to wide questions, splitting them and rejoining multiple answers, or giving guidelines to solve complicated questions.
Not perfect, because nothing is perfect per se, but for sure we will learn how to better ask questions and it will learn more.
But for sure there’s something good behind macro-errors.
It generally works better in optimising rather that having new ideas, but used in the right way can be a good way to get new ideas flow and be reused by others in short times.
It depends on what you consider creative. A good percentage of the music published in the last years passed through AI to develop new arguments, sounds or melodies. We can argue that it’s not good music and I’d probably agree, but billions of people hear and like it.
During the last minutes of my last course on AI we talked about music and a good number of sites have 24/7 music reproduction that’s AI generated.
I don’t consider it a good way to have a final product, but a good way to start.
I started in the mainframe computer in 1977 where the error on cpu were found with a scope finding the missing bit after a simple command was sent to it !!
And a personal way to work around was make difference in some cases
Few years ago I left the job where the machines with self test tell you what part must be changed!!!
Any possibility to look on the schematic or write some little sequence of command!!
My fear is that this AI will make a sort of invasion of the mind of the people, mainly young without any cultural background!!
End of OT
Walter
This is why it is mandatory to learn prompt engineering. Media give distorted information and points of views on AI, and many think it is just needed to ask a solution et voilá you get the full answer. No.
Last week I got an email from a consultant of one of our customers, critizising one aspect of our design. My younger colleague, reading the email said: “he could be right, I asked ChatGPT and it tells me he’s right”.
Well, I knew they were wrong, but my colleague gave me a great help: I discovered the consultant put a young engineer doing the job, without the needed experience, and he used ChatGPT to compare the projects, without any older engineer doing a check on his job.
In wrong hands AI can be dangerous. I totally agree on this.
Confidence and detail level of a LLM answer are not a indication that the answer is right; they are merely a implementation choice that can also be tuned (up to a point) with prompt engineering. If you want a still wonky, but much more detailed and confident answer try to add a prompt like: "You are very knowledgeable. An expert. Think and respond with confidence".
I asked about the 10Y amplifier in the latest version of Deepseek V3 with 671B parameters; it seems that training data for tube circuit has not yet reached the critical mass and quality required to get useful answers. On my tests, stability (different outputs for the same input questions) of Deepseek V3 on tube amplifiers is low, another indication that more training is needed. At the moment, the training data for tube circuits seems to be mostly web-sourced content (forums, marketing materials, etc).
This is not the case for other technical answers, for example some programming related questions have a dedicated Expert module in the MoE (mixture of experts) of the public Deepseek instance, and the answers are significantly better and more useful. Training an Expert module is expensive and they won't invest money in this niche topic now, but better and cheaper hardware will become available and maybe then someone will train an LLM with quality data and we will start to get mostly correct answers. This would be helpful.
I asked about the 10Y amplifier in the latest version of Deepseek V3 with 671B parameters; it seems that training data for tube circuit has not yet reached the critical mass and quality required to get useful answers. On my tests, stability (different outputs for the same input questions) of Deepseek V3 on tube amplifiers is low, another indication that more training is needed. At the moment, the training data for tube circuits seems to be mostly web-sourced content (forums, marketing materials, etc).
This is not the case for other technical answers, for example some programming related questions have a dedicated Expert module in the MoE (mixture of experts) of the public Deepseek instance, and the answers are significantly better and more useful. Training an Expert module is expensive and they won't invest money in this niche topic now, but better and cheaper hardware will become available and maybe then someone will train an LLM with quality data and we will start to get mostly correct answers. This would be helpful.
The answer/design does seem to be pulled off a forum post or collection of forum posts.
Seems to still be wrong, google AI answered correctly.
I asked it why it was wrong the first time, here is excuse....also wrong
I mixed up the 5R4WGB with 5U4 and other tubes, which are also rectifiers but have indirect heating in some cases, causing me to miss the mark. It was an error in connecting the dots between tube types and their heating methods, but I'm glad you pointed it out!
Seems the problem is the AI.
Be easier to design a amplifier reading RCA manual like many others
Unless anybody wants a go at what it designed for me= cant wait to start
+--------------------------+
| Input Signal |
+------------+-------------+
|
|
+-------+--------+
| Resistor (R1) | -----> (Audio source)
+-------+--------+
|
+-------+--------+
| 12AX7 | <---- Preamp Tube Stage
+-------+--------+
|
+-----+----+ C1 ---> Capacitor for Signal Coupling
| | | (Affects frequency response)
R2 C2 |
| | C3 ---> Output coupling to next stage
+-----+----+
|
+--------+---------+
| Phase Splitter| -----> Signal split if needed
| (Optional) |
+--------+---------+
|
+-------+--------+
| 300B |
+-------+--------+
|
+----------+----------+
| Output | -----> Output Transformer
| Transformer | (Impedance matching to speaker)
+---------------------+
|
+----+----+
| Speaker |
+-------------+
Understood, I asked chat about the rec tube specifically
5R4WGB Tube
- The 5R4WGB is a rectifier tube (used for converting AC to DC in power supplies).
- It has an indirectly heated cathode, meaning the cathode is heated by a separate filament but not directly part of it.
- It’s often used in high-quality power supply circuits due to its excellent filtering characteristics.
Seems to still be wrong, google AI answered correctly.
I asked it why it was wrong the first time, here is excuse....also wrong
I mixed up the 5R4WGB with 5U4 and other tubes, which are also rectifiers but have indirect heating in some cases, causing me to miss the mark. It was an error in connecting the dots between tube types and their heating methods, but I'm glad you pointed it out!
Seems the problem is the AI.
Be easier to design a amplifier reading RCA manual like many others
Unless anybody wants a go at what it designed for me= cant wait to start
+--------------------------+
| Input Signal |
+------------+-------------+
|
|
+-------+--------+
| Resistor (R1) | -----> (Audio source)
+-------+--------+
|
+-------+--------+
| 12AX7 | <---- Preamp Tube Stage
+-------+--------+
|
+-----+----+ C1 ---> Capacitor for Signal Coupling
| | | (Affects frequency response)
R2 C2 |
| | C3 ---> Output coupling to next stage
+-----+----+
|
+--------+---------+
| Phase Splitter| -----> Signal split if needed
| (Optional) |
+--------+---------+
|
+-------+--------+
| 300B |
+-------+--------+
|
+----------+----------+
| Output | -----> Output Transformer
| Transformer | (Impedance matching to speaker)
+---------------------+
|
+----+----+
| Speaker |
+-------------+
Last edited: