Electrical Engineering Stack Exchange is a question and answer site for electronics and electrical engineering professionals, students, and enthusiasts. It only takes a minute to sign up.
Sign up to join this community
I'm trying to control some powerful LED strips (about 100 W) with an ESP32 microcontroller.
To reduce cost and complexity, I'm trying to avoid:
I already have 24 V and 3.3 V on the board.
Could something like that work? I plan to dim the strip using PWM but I think that the PWM frequency should not exceed 5 kHz.
Consider that the IRLZ44N is just an example; I plan to use a MOSFET with VGS = 30 V, IS = 10 A, and Rds(on) = 3 mΩ.
First if you have a high power LED strip, it is better to split it into several segments and drive them with phase-shifted PWM signals. This makes the total current draw less choppy, which makes the ripple current lower, which makes your power supply caps happier. It also reduces flicker, and reduces the likelihood of whine at the PWM frequency from ceramic caps and power supply magnetics. And your MOSFETs will have less I2R losses. ESP32 has very nice PWM units that can do this easily.
Next you have to decide about switching speed. For EMI it's best not to switch too fast: with high gate drive current from a proper MOSFET driver it would switch in a couple tens of ns, so your wires will have square wave currents with harmonics in the hundreds of MHz, making a wideband radio jammer.
If you had 5V it would be cheaper to use a 74HCT logic chip as 3V3 to 5V logic level translator, then 5V MOSFETs.
Since you don't, you can either use 3V3 gate drive MOSFETs, driven directly by ESP32, or use a discrete driver.
The first requires the MOSFET to be in stock, which can be a problem.
A 74HCT245 as driver offers 8 channels of 3V3 to 5V translation, combined with a 5V LDO from 24V this will most likely be simpler and cheaper than discrete transistor drivers if you have several channels. Do not use 74HC, when powered from 5V, HC that needs more than 3V3 to reliably register a "1". But HCT's input levels are ideal for 3V3 logic, and it outputs 5V.
Your solution consumes 12mA through the two 1k resistors even when it's off, which is inefficient. Here's a proposal.
Note ESP32 uses a lot of current. If you are not using a switching converter to power your ESP32, a linear regulator from 24V will have huge losses. A DC-DC will be cheaper than the heatsink, not even counting the electricity.
...Previous text deleted...
EDIT: Sorry folks, Thanks AaronD. I misread the question and withdraw my previous remarks.
Get a FET with a "logic-level" gate drive. The miller plateau must be less than 3.3 V. The pull-up to 24 V should go to 3.3V. Make it small enough to use the maximum current available in the bipolar. The gate pull down should be 50k to 100k.
You can also use a 3-terminal regulator to drop the 24 v to 10 volts for the gate bias.
Gate drivers may be expensive, but they do quite a bit. If you want to do a DIY gate driver, then even the most rudimentary one will have quite a few components. But it will be very inexpensive, and the components will always be available. So - it's a tradeoff.
The circuit below is a 35ns rise/fall time gate driver for an IRF540 FET that could be suitable for your application - although any better FET would work too, of course.
PWMing LEDs directly creates high amplitude, high slew rate voltages that radiate like crazy. I suggest not doing it, and instead adding a current smoothing inductor L1, and the associated freewheeling diode D4 - it should be a Schottky type rated for 100V and the maximum LED current. The LED current will be proportional to the PWM duty cycle, with some ripple amplitude that decreases with larger values of L1.
L1's value is for 200kHz operation. For 20kHz, while retaining same ripple current, it should be 1mH. I don't suggest going lower than 20kHz - there's little advantage to it, and L1 would get quite bulky then, as you'd be looking for a 5mH, 10A inductor.
When choosing L1, ensure that the saturation current spec is at least 20% higher than the maximum LED current.
Of course, the circuit will work without L1-D4, but it will obliterate the broadcast AM band and who knows what else. The spikes will be probably picked up by a nearby oscilloscope with just a probe plugged in and laying on the bench.
R10-L10 represent the source impedance of the MCU GPIO pins and are only used in the simulation. Same goes for the load R20: you can use a dummy resistor load, but eventually you'll want to put LEDs there.
simulate this circuit – Schematic created using CircuitLab
The input logic signal is buffered by the Q1-Q2 emitter follower.
There are two control paths for the output stage Q4-Q5.
At DC: When the input signal is high or disconnected, the R1-R2-R3-R4 biasing network keeps Q5 on and Q4 off, and M1 turned off. When the input is low, Q3 is on, turns Q4 on and Q5 off.
During input edges, C1 and C2 conduct and provide fast charge injection/drainage to the bases of Q4 and Q5.
The circuit could be easily extended for synchronous rectification, i.e. having D4 replaced by a second mosfet.
The layout of this circuit is critical. A breadboard prototype should only be used for very light loads, say up to 250mA - otherwise there's good likelihood of destroying components.
The circuit put to use with a 100W LED strip must be laid out on a 2-layer PCB, with a ground plane, and with attention to all current loops - their area must be minimized, and their impedance should be kept low by using wide traces and ideally polygon fills.
One issue with this driver is that BJTs have "storage time", so they may take several microseconds to turn off when saturated. This is going to result in a few percent duty cycle distortion at 5 kHz.
Also, when the BJT turn off your driver current to turn on the MOSFET gate starts at 24 mA and ends at 12 mA. Typically, you want the drive current to be much higher (like over 1A). The rather low drive current will increase the MOSFET turn on time and result in a lot more power dissipation (probably like 50X more than with a fast driver).
