ECDL setup with custom controller

In this multi post series, I'll describe step by step series how I built by ECDL setup. Let's start with physical construction.


Design

In this build, I'll use diffraction grating to reduce the bandwidth. The grating will be operating in Littrov configuration. In this case, it's beneficial to choose a grating that has the lowest grove count while providing only single diffraction order to not lose power unnecessarily. Rearranging some equations, we find that 2nd order diffraction in grating working in Littrov mode happens at the angle arcsin(3/2*lambda*grove density). If this arcsin doesn't exist, because argument is bigger than one, then grating doesn't have higher orders. For 520 nm this critical grove density is around 1300/mm. On the other hand, we would like the grating to have the lowest possible grove density for highest dispersion and easier single mode operation. My supplier stocks only 1200 and 1800 groves/mm gratings, not 1300. And while 1200 would provide some loss in higher order diffraction it's hard to judge which one will have higher efficiency, while 1200 provides higher dispersion. For testing, I decided to order both.


Grating bandwidth

Laser diode facet size is roughly 1 μm x 5 μm and the collimator used has an effective focal length of 3mm. This means that for the feedback power to reduce by half the returned beam has to move by 0.5 μm or 2.5 μm, depending on the axis. This implies that change of angle of returning beam by just 9.5 or 47.7 milidegrees is enough to reduce feedback by half. For 1200 groves / mm grating used in this project it gives 3dB bandwidth of just 0.27nm or 1.38nm respectively.


External cavity mode spacing

When cavity round trip distance is 100 mm, then around 2e5 wave nodes fit in the resonator. Then mode spacing is 520 nm/2e5 or around 2.6 pm, much tighter than grating bandwidth. Clearly, many modes of external resonator can fit in gratings bandwidth. Fortunately, there's also an internal cavity that provides selectivity. Because the external cavity's length is defined by aluminum bracket holding parts together, it's expected to change by around 0.0023%/K. While it doesn't seem like much, it gives almost full mode hop for 1K of temperature change.


Internal cavity mode spacing

A laser diode's structure is usually a few hundred um long. For simplicity, let's assume it's 500 um long and that index of refraction is 3.5, giving 7630 wave nodes inside the resonator for a mode spacing of 77.3 pm. Overlapping both cavities and grating bandwidth creates higher selectivity filter (see picture), additionally external cavity stabilizes small drifts due to thermal changes of internal cavity. It should be noted that because of big internal cavity mode spacing, small changes to temperature or injection current will hop external cavity modes one by one retuning the system. It's therefore critical to stabilize diode temperature and current. How precisely should temperature be stabilized depends on what semiconductor is used. For eg. red laser diodes, made from AlInGaP are very sensitive to temperature, drifting 120 pm / K while free running. To maintain single external cavity mode, temperature has to be regulated to with-in 21.7 mK. For GaN diodes the requirements aren't nearly as strict.
Physical construction

Let's look from outside to inside. The entire system is placed on big slab of aluminum working as sturdy base and heatsink. On that there's Peltier module mounted, on which rests an inner baseplate. The inner baseplate holds the entire optical path. On one right there's Newport mount to which diffraction grating will be glued, on the other end there's another Peltier module, hosting brass laser diode + collimator mount. To protect the optics from dust and air currents, everything will be covered with 3d printed plastic cover with glass window positioned at brewster angle or AR coated window, depending on what will be easier to source.

Attachment 61024



And as always more details are on my blog: https://sduc6.blogspot.com/2024/03/e...er-part-1.html

wishing you luck. That’s some right tolerances you need. Looking forward to some run measurements.





QUOTE=atomd;364373]In this multi post series, I'll describe step by step series how I built by ECDL setup. Let's start with physical construction.


Design

In this build, I'll use diffraction grating to reduce the bandwidth. The grating will be operating in Littrov configuration. In this case, it's beneficial to choose a grating that has the lowest grove count while providing only single diffraction order to not lose power unnecessarily. Rearranging some equations, we find that 2nd order diffraction in grating working in Littrov mode happens at the angle arcsin(3/2*lambda*grove density). If this arcsin doesn't exist, because argument is bigger than one, then grating doesn't have higher orders. For 520 nm this critical grove density is around 1300/mm. On the other hand, we would like the grating to have the lowest possible grove density for highest dispersion and easier single mode operation. My supplier stocks only 1200 and 1800 groves/mm gratings, not 1300. And while 1200 would provide some loss in higher order diffraction it's hard to judge which one will have higher efficiency, while 1200 provides higher dispersion. For testing, I decided to order both.


Grating bandwidth

Laser diode facet size is roughly 1 μm x 5 μm and the collimator used has an effective focal length of 3mm. This means that for the feedback power to reduce by half the returned beam has to move by 0.5 μm or 2.5 μm, depending on the axis. This implies that change of angle of returning beam by just 9.5 or 47.7 milidegrees is enough to reduce feedback by half. For 1200 groves / mm grating used in this project it gives 3dB bandwidth of just 0.27nm or 1.38nm respectively.


External cavity mode spacing

When cavity round trip distance is 100 mm, then around 2e5 wave nodes fit in the resonator. Then mode spacing is 520 nm/2e5 or around 2.6 pm, much tighter than grating bandwidth. Clearly, many modes of external resonator can fit in gratings bandwidth. Fortunately, there's also an internal cavity that provides selectivity. Because the external cavity's length is defined by aluminum bracket holding parts together, it's expected to change by around 0.0023%/K. While it doesn't seem like much, it gives almost full mode hop for 1K of temperature change.


Internal cavity mode spacing

A laser diode's structure is usually a few hundred um long. For simplicity, let's assume it's 500 um long and that index of refraction is 3.5, giving 7630 wave nodes inside the resonator for a mode spacing of 77.3 pm. Overlapping both cavities and grating bandwidth creates higher selectivity filter (see picture), additionally external cavity stabilizes small drifts due to thermal changes of internal cavity. It should be noted that because of big internal cavity mode spacing, small changes to temperature or injection current will hop external cavity modes one by one retuning the system. It's therefore critical to stabilize diode temperature and current. How precisely should temperature be stabilized depends on what semiconductor is used. For eg. red laser diodes, made from AlInGaP are very sensitive to temperature, drifting 120 pm / K while free running. To maintain single external cavity mode, temperature has to be regulated to with-in 21.7 mK. For GaN diodes the requirements aren't nearly as strict.
Physical construction

Let's look from outside to inside. The entire system is placed on big slab of aluminum working as sturdy base and heatsink. On that there's Peltier module mounted, on which rests an inner baseplate. The inner baseplate holds the entire optical path. On one right there's Newport mount to which diffraction grating will be glued, on the other end there's another Peltier module, hosting brass laser diode + collimator mount. To protect the optics from dust and air currents, everything will be covered with 3d printed plastic cover with glass window positioned at brewster angle or AR coated window, depending on what will be easier to source.

Attachment 61024



And as always more details are on my blog: https://sduc6.blogspot.com/2024/03/e...er-part-1.html[/QUOTE]

Let's talk more about the controller. After a lot of code writing, I managed to power up almost everything and even implement the simplest possible control loop. Everything is controllable over USB with VISA and all settings are persistently stored in flash. All the bookkeeping (usb, interrupts handling, fault monitoring, etc) is running on core 0 with core 1 fully dedicated to (future) control loops.

All analogue hardware is fully functional, though with slight changes from the original:

• the MAX6035 reference source has been substituted for a cheap 5V regulator. It turns out that thermistors' analogue chain have enough PSRR to work properly even with not as well regulated ref voltage, saving around $3 per board. The slight downside is that laser diode accuracy dropped slightly.
• Feedback resistors around op-amp had to be increased to increase noise level to ~1LSB
• 1500uF bulk capacitance had to be added on power rail to absorb reactive power from switching TEC outputs quickly.

And now what everyone is waiting for - real world performance measurements:

• The laser diode constant current source is quite bad with 0-100mA range, 0.2mA precision and accuracy of around 2mA. That was expected as I don't need to set specific current often, but I need to adjust it quite precisely to hit centre of SM operation
• The thermistor paths at the beginning had true 16 bit performance, but that makes averaging ineffective, so I increased noise until the histogram showed a standard deviation of 1.1LSB. The ADCs run at 50ksps and are downsampled 512 times, theoretically giving another 4 bit resolution boost. With current analogue gain, the resolution hovers between 0.4 mK and 0.6 mK in >40 K range. The accuracy is bad, around +-1K but that's not important for ECDL and can be compensated using 4 point calibration
• The Peltier modules are driven with a combination of PWM and sigma delta modulation. To get a good filtering, PWM has to run above 50 kHz, limiting resolution to around 8 bits. On the other hand, sigma-delta modulation requires a lot of transitions that generate too much switching loss. The solution used here is PWM that is switching between two values every period, with switching distribution selected by sigma-delta. It's not an optimal solution for minimizing in-band noise for a given set of edges at a given clock speed but it's quite close, and it's computationally efficient thanks to DMA

In the steady state, the system wanders around 1mK on both temperature sensors. When I've put ice cubes on outer baseplate the inner baseplate temp changed by 100mK max and LD temps changed by 20 mK max (with barely tuned PID loops).