Power integrity design review of LED driver and audio amplifier PCB

[sephalon]

I am seeking advice on improving the power integrity of a PCB featuring a HT16K33 driving 6 segment LEDs and a YX8002 (equivalent to LM4871) audio amplifier.

Here are the relevant sections from the schematic, but I would like to directly refer to the KiCad project for the PCB design available on GitLab.





The circuit draws around 150 mA at 5 V when idle (i.e. with the segment LEDs turned on and the speaker muted). I tried different linear bench power supplies and even the 5 V rail of an ATX power supply, but to a variing degree, I experience issues with power integrity as illustrated by the following scope screenshots.

The 5 V power rail drops by up to 500 mV while the LED driver is multiplexing through the LEDs; in worst case, when all LEDs are driven in 1 Hz blink mode, I can hear a hissing sound, presumably coming from the power supply:


When the amplifier is turned on as well, the power rail is distorted even further and the LEDs become visibly dimmer during the loudest parts of the audio sample:


Decoupling capacitors C7 and C16 are placed physically as close as possible to the respective power pins and are connected to the ground plane (IC pin – cap pin – via pin). Besides increasing C16 from 100 nF to 10 uF (with no noticeable effect), I tried to add several different electrolytic capacitors in the range of 10s and 100s of uF in parallel to C16 (admittedly in a crude way, by pressing the capacitor leads against the power and ground pins of U5, as I had not planned for an additional decoupling capacitor footprint). This smooths down the distortion somewhat (in the sense that it always looks more like the second scope screenshot with the audio amplifier being turned on), but the same distortion pattern is still visible and the user-facing issues (dimmed LEDs and PSU hissing) do not go away. Hence, I conclude this is not really a decoupling issue.

With the HT16K33 sinking around 30 mA per row and a maximum of 10 segments connected per column, the display would draw around 300 mA worst-case. The audio amplifier draws up to 600 mA (but of course I do not drive it to the maximum at all). With my power supply net having a total length of 30 cm alone, a trace resistance calculator tells me that I would have a voltage drop of several hundred mV with my trace width of 0.25 mm used for everything – so in my current understanding, power-carrying traces being too thin is the main issue for the massive power rail voltage drop seen here.

I have now doubled the trace width of the 5 V power rail and the common LED driver traces to 0.5 mm (see KiCad project on GitLab). Do you agree that is the main power integrity issue of my design or is it something else? I am also grateful for any other hints to improve my design. Thanks in advance!

[ArdWar]

Quick and dirty snapshot of your PCB with +5V net highlighted before someone gives you 10 pages worth of lectures about common document interchange format


Well, there's your problem
At a glance:

• It's way too thin for a power trace. Make it much thicker trace, or make it a large polygon pour, or better yet make the entire layer a power plane
• It went through way too many vias. For your particular design it should be other (low speed signal) traces that gives way to the +5 trace, or use multiple vias in parallel if layer weaving is unavoidable
• Big power consumers are generally way too far from the supply input. Your amplifier is basically as far as it can be from the power pins. It will be much better if you rearrange your placements in such way that your main power consumers is much closer to the power input
• All your components are basically powered "in series" to each other. You can try using dedicated trace from power input pins to each bigger power consumer (ex. one trace for the amps, one trace for the LED driver, one trace for everything else)

[MarkT]

I have now doubled the trace width of the 5 V power rail and the common LED driver traces to 0.5 mm

Well try more like 2.5mm and you might be getting somewhere.  10 cm of 0.25mm track has 0.2 ohms resistance when cold, and somewhat more when hot (carrying lots of current).

Standard PCB (1 oz copper) has 0.5 milliohms-per-square, so you can simply divide the trace length by its width and multiply by 0.0005.  So (100mm/0.25mm)*0.0005 = 0.2 ohms.  (100mm/2.5mm)*0.0005 = 20 milliohms - which drops 20 mV per amp, much more like it.

Power traces also should be wide for low inductance, whether they carry large currents or not.  1mm is a good starting point just from that perspective.

[sephalon]

I have increased the power trace width, removed all vias and separated the power path for the LED driver and amplifier. A new PCB is in production now and I hope it will show much better power integrity results. Thanks for your support!

[sephalon]

So I have received above new PCB (also pushed to GitLab), with 1 mm power traces, no vias on the power rails and a separate power path for the audio amplifier. Unfortunately the power distortion due to the LED driver looks pretty much the same, with virtually no improvement whatsoever. I did not assemble the audio amplifier yet.

To rule out that the power trace is still too thin, I soldered wires directly onto both ends of the decoupling capacitor C16 and connected them to the power supply. This reduces the voltage dropout by about 100 mV, but still leaves us with drops of up to 300-400 mV during switching cycles.

I also noticed that the LED driver emits quite a lot of noise in the hearable audio range; especially when set to blinking mode, this would already bother me as a user.

Any further ideas how I could improve the situation? Is the HT16K33 an inferior LED driver (at least for this task) and should I replace it with something else?

[selcuk]

What is the regulator which outputs 5V in the circuit? Can you share the power supply part of the schematic? I recommend you to share pdf of the schematic along with the project files to reach more people.

What is the measurement point for those oscilloscope images? Did you measure it at the regulator output or at the led driver supply pins?

[sephalon]

What is the regulator which outputs 5V in the circuit? Can you share the power supply part of the schematic? I recommend you to share pdf of the schematic along with the project files to reach more people.

I have not designed the power supply for production use yet; so far I have powered the circuit with two different linear bench power supplies (rated for 1.5 A) and via USB from my laptop. The first revision of the PCB with insufficient power trace width I had also hooked up to an ATX power supply (with not much improvement).

What is the measurement point for those oscilloscope images? Did you measure it at the regulator output or at the led driver supply pins?

I have measured both at the power supply as well as the decoupling capacitor and supply pins of the LED driver, without seeing much of a difference…

[selcuk]

I guess there is a cable connection from power supply to your circuit. But it may not be related to this connection's resistance or inductance if you are measuring the same signal on the power supply output. In this case you need to supply the fast current needs through local capacitors on your board.

You've already increased the capacitance for HT16K33 and amplifier. Either those additional capacitors have high ESR or have leads that introduce inductance. I recommend you to add additional bulk capacitors on board. These should be SMD and low ESR. The track should go under capacitor pins and then go to HT16K33. You don't need to order new PCBs. You may try to scratch solder mask of 5V line and GND nearby to place SMD capacitors. You can try a few 22uF, 47uF MLCC or tantalum low ESR capacitors. Use 100nF for C16 and 1uF for C7, since they are recommended. You will add additional capacity to them.

You may try to run amplifier and led driver separately to bound the issue to a smaller area.

[PCB.Wiz]

To rule out that the power trace is still too thin, I soldered wires directly onto both ends of the decoupling capacitor C16 and connected them to the power supply. This reduces the voltage dropout by about 100 mV, but still leaves us with drops of up to 300-400 mV during switching cycles.

That shows you that lead resistance is important.
How long and what gauge  are the leads to your power supply ?

I also noticed that the LED driver emits quite a lot of noise in the hearable audio range; especially when set to blinking mode, this would already bother me as a user.

What does that even mean ? A driver does not emit noise.
Do you mean the Audio amplifier has cross talk from the LED driver, which you can hear in the speaker ?

Any further ideas how I could improve the situation? Is the HT16K33 an inferior LED driver (at least for this task) and should I replace it with something else?

If you are trying to mix Audio amplifier and LED drivers, you might want to look for a LED driver that has linear current control. Avoid one with digital dimming.
The PCB design should also totally separate Audio and LED, like you are making two boards.

At a glance, the circuit does not make much sense.
The HT16K33A has current source ROW and strong N-FET sink COL, and they show Anode to Row and cathode to COL as expected,  yet you have wired backwards on SCH ?
If you are using brightness control/dimming, the HT16D33B looks to have 16 steps of Analog segment dimming.