"Think Digital, not analog"
This is a continuation of the 1000W class D amp design. It may be a good idea to read that article first, as this is a highly modified version of the previous design and explains many of the reasons behind my design choices.
So far, I have gone through 4 Revs for this project:
- Rev 1 (previous PCB) - non-functional 8/26/2025
- Rev 2 (Partially functioning PCB) - December 2025
- Rev 2.5 (small board to verify layout and schematic of TAS5830) - finished 1/11/2026
- Rev 3 (Final PCB) - finished 1/25/2026
Compared to my previous design (rev 1), the changes are as follows:
Output + DAC: Instead of using a PCM5242 + TPA3255, I replaced it with a single TAS5830 IC. This takes in a digital I2S input. I love this chip for many reasons:
- Simplicity: (It's a DAC, Amp power stage, DSP, and digital logic, all in one IC). No external circuit for power-on/off pop prevention (Update: there is a slight tick when the amps are turned on, but it's not loud). Much fewer external components are needed. No more concerns about grounding: the Amp IC is fully digital and uses a single ground, as shown in the EVM.
- Performance: Extremely low noise (40uV - Can barely hear the hiss of my JBL Control 25-1 speakers when I place my ear up to the tweeter. Lower idle current draw due to 1PSW modulation compared to BD (TPA3255). The heatsink and inductors are barely warm after blasting full volume for a while, remarkable!
- Manufacturable: Not QFN! Assembling QFNs is really difficult for me. A TSSOP component takes me at most 5 minutes, while the QFN package for the PCM5242 took my friend and me over 1 hour and still didn't work! I had to redo the QFN packages on my previous design numerous times in an attempt to get it to work, wasting time and energy.
- Validated: SW mode is very powerful and allows for diagnostics, config, and even level reading (as well as DSP onboard). A similar IC (TAS5828) is validated with the ADAU1452. Since the TAS5830 shares the same pinout and registers as the TAS5828, it will work with the ADI DSP. Here is the source of validation: https://oshwhub.com/lbl66666/adau1452-tas5828-gong-fang
ADC IC change + modular approach: In the final design, I made a external board that allows me to swap the ADC for another board with a completely different chip. Or I could even create a custom board for something like a Pi Zero 2W to do networked audio (Ravenna) or streaming. However, I did switch the ADC to a PCM1863. It supports software control and offers a higher SNR than the PCM1802. The IC also provides PGA gain control without external op-amps, in addition to differential inputs. I also used an ultra-low-noise regulator because the previous one was oscillating (AMS1117-5V). Don't use that regulator! I used a TPS73633DBV for the digital side regulation and ADPL44002AUJZ-3.3-R7 for the analog side regulation for the absolute lowest ripple and noise (also highest PSRR)
Change in components used:
I do prefer to use larger package sizes, but there is a critical issue with that: inductance. It correlates to component body size. Unfortunately, 0402 components are the best for decoupling this amplifier, as the frequency at which the component acts like an inductor is pushed higher than larger packages. However, I find it hard to do 0402 well, so I made the trade-off of using 0603, which is acceptable. This applies to smaller capacitors, but I still choose to use larger components (like 1206) for high-voltage, high-value decoupling.
Added a MCU: I used an RP2350 for the MCU to control everything. I used one core for the main controls and one for fan control. It communicates with a UART display and sends commands to it to change pages or to display text
Now, here are the disadvantages of this new design:
- Power output is reduced from 140W to 100W for THD+N=1% measurements
- Each channel will need its own IC (In PBTL) to output 100W at 4R (29V), increasing IC count
- Channel count is reduced from 6 to 4
- SW mode of both amp and ADC IC adds complexity from the software standpoint (MCU)
Since Rev 2 was a long story, I wrote a separate article about it. I would talk about the boards related to my current rev in this article and the things I learned when working with this IC below.
Link to site discussing rev 2
The following discusses Rev 2.5 and Rev 3:
The issues with Rev 2:
Wrong MCU pinout:
The pinout for the MCU in KiCad is completely wrong, resulting in numerous issues. I had to desolder it and connect external wires to the board.
The use of multiple I2C buses
It is better to use a single I2C bus rather than multiple if possible.
No sound from TAS5830, reliability issues
This was caused by a couple of issues. First off, I misconnected some wires! DVDD should be receiving 3.3V, and in my initial design, it's not connected to anything! What happened instead was that the gate drive (GVDD) was accidentally swapped with DVDD, which caused issues like 5V on the 3.3V plane (very bad), and the digital side not receiving anything. Of course, it shouldn't work at all, but that's why I had horrible sounds (and pops) from the speakers, no music.
Fixing this made the system to finally play music for the first time. However, when recycling power for both the MCU and the board, the system did not detect the I2C device. I suspect a couple issues causing this problem.
Improper decoupling:
This one is bigger than I thought. In the datasheet, it shows a example layout. I did not notice how close the decoupling capacitors were to the IC at first, until I overlaid the front and back layers:
This is how close the decoupling capacitors need to be, to the point of causing design rule errors!
I then noticed how far my decoupling capacitors were from my amp IC. The datasheet explicitly said to place the capacitors no further from what's shown in the example layout to prevent damage.
I initially thought that the amp will only self destruct at high volume for that problem, but since class D amplifiers work like a PWM modulator (switches on and off an output at very high frequencies, avoiding the transition between the on and off states), the Di/Dt is just as high. I suspect that the output stage is damaged as it reaches voltages above the absolute maximum rating - killing the chips.
Bootstrap capacitor issues:
I suspected that bootstrap capacitors could be a cause of the issue because of different characteristics. I decided in the end to use the same TDK capacitors as used in the EVM. This is because when I swapped the Samsung capacitors from my board to the EVM (while testing faults), the EVM stopped working properly. The sound was really distorted and not like music I was playing at all, Swapping back the TDK capacitors back to the EVM made the circuit work properly again.
Even if the issue was a soldering issue and not specifically with the Samsungs, I think the bootstrap capacitors are a really crucial component to a class D amplifier. The parameters between the Samsung and TDK capacitors (such as SRF, DCR, inductance) can be really different. Thus I decided to use the same bootstrap capacitors in the end.
While flipping back and forth between my design and the one of the EVM, I noticed that they had a TI TPS linear regulator supplying 3.3V for the amp ICs. While this may not directly cause issues, the linear regulator used was low noise (thus low ripple). This means that a really stable and clean supply was used to supply the DVDD pin.
In contrast, I used a switching regulator on my board. Some switching regulators can emit transients larger than the rated output voltage when under light load. Maybe the regulator destroyed the IC by supplying more 3.3V, thus killing the digital logic part of the IC?
In addition, I also had the catastrophic event of the IC feeding back 5V to the 3.3V line due to a mis- wired pin. While this is caused by a mistake on my side, I still believe it's better to still isolate the power bus to prevent cascading failures in the future.
Lack of pullups for the GPIO
Deviation from the EVM
The fixes to the issues:
Improved decoupling for PVDD:
Following both the EVM and datasheet design schematic as close as possible, I followed the same geometry for the top and bottom copper pours for the PVDD net. I placed the decoupling capacitors and vias as closely as I can to the design provided. However, I changed to 0402 decoupling caps to 0603 0.1uF, and from 0805 to 1206 for 10uF. I changed the bulk decoupling capacitors from SMT to through hole and increased capacitance from 390uF and 470uF (680uF for final design)
A different type of decoupling capacitor for PVDD:
So while searching for a small and high performance capacitor for my PCB, I noticed that the types of capacitors I want (10mm diameter, high capacitance, less than 22mm tall) does not really exist in the 10mm diameter. I then stumbled upon solid state (polymer) capacitors. These things offer much better ESR and current specs than the original electrolytic capacitors used on the EVM. I believe that this will lead to better transient response to responding to the high di/dt of the amplifier IC, reducing ground bounce and to provide better power filtering for the amp IC too.
Dangerously close design margins in the EVM's design:
One major reason for the changes from 0805 to 1206 was because the EVM's designer decided to use 35V decoupling capacitors for something that should be able to handle up to 30V. While this may not immediately pose a issue, this puts the decoupling capacitors under more stress as they are ran closer to their maximum ratings, shortening lifespan. Because of that, I decided to use 50V rated capacitors for the larger decoupling and bulk caps, with better or similar specifications.
In addition, smaller MLCC of larger value will lose capacitance with DC bias. Therefore the 1206 capacitors will exhibit better performance as the capacitors should loose less voltage given the same DC bias. So this change should only improve decoupling performance.
Output LC filter changed - from 2220 to 1210 (EVM used 0805) used a Samsung capacitor model similar to one specified for a TPA3244 sound bar design.
I used the same 0.68uF capacitance, but rated up to 100V for the sake of audio performance (wanted to use the same capacitor as the EVM, but the capacitance decreases so unbelievably much as the voltage approached 30V - degrading sound quality). Ceramic capacitors are known for bad sound as the change in capacitance with voltage bias causes distortion, reducing transparency and sound quality. Although the amplifier is a closed loop architecture, I believe a better filter network will still result in a better sound as maybe the closed loop network can do less work (resulting in less artifacts) or help out with the part of the frequency response that the IC is unable to compensate for. Sadly, I can only guess because the information of how the closed loop architecture is not listed from TI, though it is listed to compensate for the non linearities on the output. I Placed the capacitor as close to the inductor hopefully to reduce faults caused by the closed loop amplifier (lowering inductance).
I replaced the decoupling capacitors on the low power side with the same ones as used on the EVM (0603). The smaller package allowed me to place it as close as I can to the IC (despite this change making it harder for me to solder).
And I also access to all GPIO pins for debugging, with the required pullups that I omitted for rev 2:
End results? It worked! It didn't just work, it worked really well! Gone are the reliability issues and hello to some amazing sound from a class D amplifier! This thing passed my validation tests and never had issues again.
Comments
Post a Comment