I've built a level converter for a digital PWM signal that runs at 800KHz. The circuit goes like this:
*I'm using a BS170 transistor instead
The problem is the signal gets distorted at around 10KHz, I think the transistor is supposed to handle the bandwidth. How can I raise the frequency limit?
Is the distortion on the rising edge? Pull-up of Q1 \$V_{\text{ds}}\$ through R2 is a limiter for bandwidth. At close to \$V_{\text{ds}}\$=0V, BS170 has \$C_{\text{ds}}\$~40pF and \$C_{\text{gd}}\$~20pF. So, time constant for R2(\$C_{\text{ds}}\$+\$C_{\text{gd}}\$) ~ 4usec. After the rise of GPIO has played out, the last 2V to 2.5V of rise will follow that 4usec time constant, which makes the large signal bandwidth less than 100kHz. Other capacitance in the circuit will make things worse. Lowering R2 to 1kOhm would help get the bandwidth close to 800kHz.
Should also point out that \$V_{\text{th}}\$ of the BS170 can be as high as 3V, so there could be units that won't properly turn on. A lower voltage part with lower \$V_{\text{th}}\$ and smaller die size, to reduce part capacitance would be a better choice.
Put a capacitor from source to drain. Start small (say, 22pF) and observe the effect. Try different values. Smaller is better as long as it works.
This provides some AC coupling to boost bandwidth, but still allows isolation at DC.
I did not look up the specs on the BS170, and am not familiar with it. I have done this in the past with a BSS138 (if I remember correctly), which is a good choice. It is a "jelly bean" which is available from many vendors and at low price. The 2N7000 has a higher Vgs(th) than the BSS138.
