A tiny, ultra low profile wideband spectrometer for everyday needs (test lamps for color temperature & CRI, UV-A /UV-B dosimetry, ...)
To make the experience fit your profile, pick a username and tell us what interests you.
basic demo for the Stellaris Launchpad (LM4F120H5QR) that streams rows of tab separated integer values. Use With Energia. Latest version on github!
ino - 9.99 kB - 01/02/2017 at 16:58
Here's an inverse solution - when detecting in a wavelength range where LEDs are the go-to light sources and consequently narrow-band, multispectral sensing is easier to do, and the spectral resolution is more sharply defined.
This goes both ways though, as a set of LED dies needs to be available for both photodiode-mode and emitting diodes.
Sometimes I find I entertaining how ideas go full circle. As I was researching some other optical principle, these "multichannel optodetectors" crossed my way. They're effectively what I was striving for with the Optical inch project, albeit at a more reasonable price point than these hermetically sealed optical hybrid modules ;)
Meanwhile I'll keep claiming the approach with LEDs while it lasts.
40x40 mm² stainess square tubing with 8mm tapped holes - one for mounting an aluminium cube, the other to pass the "light" through and the PIN diode array PCB glued to the cube. Add a 1.5x3 mm² pinhole made from 2mm thick lead sheet, light seal all openings with copper tape, Al tape over the pinhole. Done. Note: this is not a transmission measurement but a backscattering geometry.
horziontal: time (50ms intervals), vertical: 7 of 8 channels (something went wrong, maybe a connector came off...)
What you see here is the raw 1D image of an 8 pixel x-ray pinhole camera observing a piece of stainless steel illuminated by an x-ray tube ~8cm away as I'm waving it around (duct-taped to a PP tube and held where the dose rate is <2µSv/h).
...And here's just the channels with a bit of smooth kdensity. At approx. sample #610 the x-ray supply is switched off.
It has proven effective to let the analog ground float inside the metal enclosure and connect case GND to the digital ground of the launchpad.
Let me conclude this log entry with a snapshot of the pinhole camera setup. Note this is a different board but I still put that Faraday shield over the exposed input pads. Black cable is chassis GND.
edit: Settings: 18kV, 110µA. Nice.
"I still have the parts for your DDC118 booster pack, do you have any assembly instructions?"
"throw that old PCB away"
Here's the new one. Guess I'll find out tomorrow that those u.fl connectors are spaced too closely or that the diff pairs have a polarity twist somewhere.
new minor revision:
- has proper footprints for U.FL connectors. They feel right for the job, right now I don't care much about the added capacitance due to coaxial cable
- tidies up passives
- actually spells out the channel numbers at the connectors and signal names near the pin headers ;-)
- removes LDO
- introduces differential DIN / DOUT
- introduces series termination resistors
- has 2x1 receptacle and 2x2 pin header for stacked daisychaining (first board has GND jumpers, all except for the base board omit R5/R6 in favor of a pair of wires that re-route DOUT/!DOUT, last board is unmodified (not pretty, but I couldn't be bothered to go 4-layer at this point
- introduces NC7SZ175 SC70-6 single UHS D-type flipflop which synchronizes CONV transitions to CLK_4X rising edge within < 5ns as per requirement (ds: sub-10ns for best performance), since I don't want to ensure these stay synchronized no matter what the Timer/Counter configuration I'll sync it in hardware. Will have to investigate beating anyway (+/- 0.5 clock cycles = 200-250ns w.r.t. 100µs-10ms integration time...)
- not yet given up on shields
- short circuits at the corners to avoid power-up when board is in wrong orientation (avoids killing the expensive IC)
- see if moving SPI clock (not in sync with converter) up past the REF3040 (U1) changed anything
- not too excited about those digital traces running across the reference buffer. DOUT/!DOUT now diff routed to cancel capacitively coupled possibly async (w.r.t. ADC clock) bitstream, async SPI clock trace relocated, rest of the signals are static
- build 2 new boards and use a full 16 pixel sensor. Oh boy...
edit: had a few hours of sleep, found mistakes and fixed them ;-)
What's more important - achieving a goal or seeing a project through? You decide. ESPROS obviously managed to get a simple (diffraction based?) filter matrix slapped onto a camera sensor, making a cost effective and elegant device called Viavi.
Need UV A/B as well? Grab a VEML6075.
Just found this very interesting simple-ish NIR spectroscopic fingerprinting sensor which goes in the general direction of the Optical Inch. Same class of device although the impelementation is quite different because it uses dielectric narrowband filters.
"The AS7263 is a digital 6-channel spectrometer for spectral identification in the near IR (NIR) light wavelengths. AS7263 consists of 6 independent optical filters whose spectral response is defined in the NIR wavelengths from approximately 600nm to 870nm with full-width half-max (FWHM) of 20nm.
An integrated LED driver with programmable current is provided for electronic shutter applications.
The AS7263 integrates Gaussian filters into standard CMOS silicon via Nano-optic deposited interference filter technology and is packaged an LGA package that provides a built in aperture to control the light entering the sensor array. Control and Spectral data access is implemented through either
the I²C register set, or with a high level AT Spectral Command set via a serial UART."
In the previous log it was shown that half of the ADC turned out to be non-functional. Luckily it wasn't the last board so another one was grabbed.
A few hours later we have a modified PCB with 1.27mm headers (perf board circles cut in half to yield 1.27x2.54 grid pads
and thanks to quick delivery from China also a big pack of u.fl cables and connectors (sculpting the pads-to-be for U.FL was something that made me want to redo the PCB - Rev1E PCB on the right is for reference so you can see the GND polygon)
Here's the fun part: soldering the U.FL cables and attaching 0.635mm wires to route the connections to the diodes
and it's done. Let's mount the PIN diode array (it's hard to see but there are 2 diodes on each of the 8 chips, 16 pixels in total, only the upper half is in use right now)
I haven't bothered to get the pinout right so the channels are out of order but you can convince yourself that
* all channels are working to spec
* no shorts
* gains are compatible
test setup: reset and acquire background for dimmed laptop monitor showing a black image with a 1px white line, move diode array across the screen, illuminating one pixel at a time as they pass. Note also that the zero line is noisy due to brightness fluctuations in the backlight an that the LCD is not completely dark, so some jaggedness is to be expected.
coming up next:
* electrically shielded housing, provisions for x-ray aperture
* systematically redo PCB mods for noise mitigation and quantify significance
* measure some stuff!
You may have noted that we're diverting a bit here. That's be cause the hardware developed here will be used in an x-ray backscatter imaging experiment. Apart from that we're still on track with the Optical Inch.
the DDC acquisition chips achieve 100% sampling time by alternating between two integration stages per channel identified by "Side A" and "Side B" for the two integration phases.
Now these integrators are not made equal, which causes different offsets and scaling factors to be applied to the acquisitions. I did find the offset issue somewhat disturbing so I addressed it first.
Let it also be noted that the CONV synchronous effect might also be in part due to LED activity or the CONV signal / trace itself. I remember it getting worse as the range is decreased.
Here we go:
unfortunately the first four data channels (mapped to input channels 1, 3, 5, 7) have some issues. I've contacted Ti to see if they have an idea and can help me out.
The remaining four channels however look quite promising, they still work quite well. For the DDC118 with 8 inputs there are 16 offsets to be kept track of. Here's what the output data looks like without (and below, with) Side A/B offset compensation:
this is 500 samples/sec, 350pC full scale and the board has just a bunch of 20cm long coaxial cables attached to the inputs.
I've also created a github repository and uploaded the demo code.
1. 50 Hz noise still can easily couple into the system
2. LEDs with a larger bandgap (for 400nm, off by 0.3eV) can still detect some 450nm light from white LEDs
3. there's other noise present, possibly because of the USB 5.0V power, microcontroller and UART activity - and it's correlated among the channels
200 samples/sec, 200 pC range:
plot for [i=2:9]'20170102-16 capture20.txt' u ($0/200):i smooth kdensity bandwidth 0.01 t sprintf("ch%d",i-2)
Alright, this project definitely got some love today. That PCB is really not that great (as in: damn that DDC118 is super sensitive).
Changes (3 years in the making ^^):
- add missing bypass caps, mostly 220nF ones, test direct bypass to AGND polygon (no significant change from the latter)
- insert 100R clock termination resistors for T2CCP1 (@2.5 MHz) and DCLK (@10 MHz)
- insert 22R in the trace supplying the voltage reference -> no significant change (not shown in layout changes)
- insert 10k series resistance on REF3040 output -> no significant change (not shown in layout changes)
- omit 74AHC1G14 single schmitt inverter. I had re-routed the cap next to it for proper 3.3V bypass including 10R resistor to the caps below but no improvement was evident. A differential clock source only really shines on larger PCBs and flex cables.
- see that cap between the second trace on the top and the LDO tab pad where it says "wrong GND point"? That cap was moved upwards because connecting it to the bypass caps behind the filter would only couple noise from the input to the output. The two traces on the top right are static and don't couple much noise into the rails.
- add EMI shielding (tin coated sheet metal cut to size) above the chip and sensitive traces. That did away with a good fraction of the 50 Hz line perturbations which are superbly evident at 100 samples/sec. The A (B) side samples themselves are supposed to be rather consistent but showed slow beating without the shield. Additionally the noise introduced by an approaching hand is mostly gone without anything connected to the inputs.
- endianness issue resolved, both 16 bit and 20bit modes look good
- right now both integrator sides are being printed, looking good at 400 samples/sec but that might get worse as integration time is increased due to differences in the capacitances and active analog front end circuitry
- SPI speed changed to 10 MHz
Next: There's a hint regarding noise performance. For best performance the CONV transition needs to happen within +/-10ns around a T2CCP1 rising edge. Right now CONV is generated from within an ISR which has delay and jitter.
This may possibly be solved by routing a trace from a T1CCPx pin to CONV and subsequent configuration of both timers. All timers have pre-scalers and can be synchronized to some extent. Not sure what cases can crop up though. Maybe It's best to latch CONV with T2CCP1 and keep the interrupt based scheme.
NC7SZ175P6X looks right for the job.
In summary, ditching the schmitt inverter for !DCLK (which likely inejcted more noise than help reject transients) and adding the E field shielding did a lot to improve performance whereas it is hard to tell apart what that course of ground re-routing and bypassing actually did. Maybe it wasn't that big a deal to begin with. Powering the Launchpad + DDC118 externally and testing it with floating pot. laptop showed no difference.
Become a member to follow this project and never miss any updates