The Probe-Scope is an open source 60MHz 250Msps oscilloscope probe with all guts built right in, that plugs directly into your PC via USB
We've squashed the bugs and straightened out the kinks... Here's the first screenshot of real data coming in from the ADC! We use pyqtgraph for the data rendering, which performs excellently and has all the plotting and UI features we want baked right in, so we've avoided sinking a ton of time into displaying the waveform. Tomorrow we will add the actual units to the GUI and support setting the parameters of the device, and on Friday we'll have the real boards in our hands.
(disclaimer: we're driving the ADC directly with a function generator here, just for simplicity's sake)
(also disclaimer: we're not using a teensy and I don't know why Windows thinks it is one, only on this computer does it say that...)
As we mentioned a few weeks ago, we had hoped to post an update about the software by now. So where's it at? We haven't shared anything yet because we have a lot of moving pieces in this system (FPGA acquisition, FIFO, and sample output, PIC sample ingress, PIC USB comms and protocol, host comms and protocol, host GUI, and host measurement features), many of which are working quite well in and of themselves, but the product as a whole isn't quite ready to roll. We will be finishing it this week of course, and we're confident we will have something impressive to show off by Saturday.
We haven't been doing nothing, however: Aside from continually working on the software, we put together a demonstration unit and showed it off at the Microchip MASTERs conference about two weeks ago, along with a very early version of the software. Check it out:
And while it's not really relevant to what we're doing now, here's a neat short video I put together a while ago of us assembling the digital test board:
Finally, we just got word (and pictures!) from our PCBA vendor that our final boards have been assembled, so we'll get those in the next two or three days, just in time to show them off to you all in our final submission. Stay tuned!
We recently finished the layout of our final probe scope available on our github, now an obligatory PCB picture:
Here is an animation of how the Probe Scope stacks, to allow expanding to multiple channels all connected to one PC. The contacts on every Probe Scope distribute the trigger signal from the master Probe Scope to all the slave devices for synchronization.
We sent away the PCBs to our favorite PCB vender, PCBWay for assemble and fabrication.
In other news, we are working hard on the Digital Test Board, developing all the software. We hope to post an update at the beginning of next week.
We recently got our PCBs in and built up, and having detailed the testing of the analog boards, it’s time to show off the Digital Test Board in its early stages. Unlike the analog boards, where the bulk of the work was done in designing them before they were made, this was straightforward to design and will take much more effort to program. So, expect some more updates over the weeks as all the software and firmware gets put together.
First, a glamour shot:
We need to get the final Probe-Scope board designed and sent out fairly soon, since the lead times for getting it made and assembled (and we will be getting it assembled) are pretty long due to the high density, and the competition deadline is approaching. So, to start off, we’re just verifying that all the pieces of the digital test board work so that we can be comfortable reusing those schematics in the final design. That means probing the ADC to make sure it’s outputting data (it is), loading code onto the PIC and FPGA to make sure they run with the correct clock sources, and then eventually (not done yet but soon) implementing a simple USB device on the PIC, and operating the HyperRAM with the FPGA.
The PIC was easy to get running as we have some lightweight test code we often reuse, and it works just fine with the external 24MHz oscillator (which is basically required for USB to work). The FPGA, on the other hand, was a bit more of a challenge for me as I have not used Lattice Diamond before, and its documentation is very spread out. Pro-tip: the file list has tabs at the bottom… you really want to check out the process tab. At any rate, I got that running too, clocked off the output of the ADC. Watch it all blink:
This leaves the USB and the RAM to be tested. The USB connections and clock, I’m not too worried about, they match known good designs. The RAM however, I’m also not too worried about, but it remains to be proven that our concept of clocking it at a higher speed using the FPGA’s PLL will work in this exact configuration. Stay tuned…
EDIT: The PLL for the RAM worked perfectly fine, thanks to the configurator it wasn't a pain at all.
Having recently received the parts and PCBs for all of our subsystem test boards, this post is one in a series detailing the buildup and testing of those boards. Today we’ll be showing off our Analog Frontend board, which includes the Variable Gain Amplifier (VGA) and Filter that makes the whole thing work. To start, here’s a picture of the finished board
After cursing the designer of the board (me), who decided that 0402 was the right size for most of the components, I finally got it assembled and up and running. One of the first things that I noticed was that it ran hot. In the back of my mind, I knew that the VGA (an ADRF6518) drew a lot of power (400mA at 3.3V or around 1.3W), but I didn’t quite consider the thermal implications of that. Checking the board with a thermal camera (shown below), the VGA got all the way up to around 70°C, which is actually right about the calculated value, but of course, I forgot to calculate it beforehand…
The first thing I did was write some Python scripts for my Raspberry Pi to control the ADRF6518 (over SPI), letting me play with its bandwidth and gain settings, along with controlling the MCP4728 DAC that set the Offset and Gain levels. Once I got it all set up and working, I connected it to our DG8SAQ VNA in order to get some frequency response plots, and they looked amazing! Here’s a picture of the board all hooked up for testing. I’m using a Tek P6248 differential probe (plugged in on the right) to feed the signal into the DG8SAQ.
Shown below is a screenshot of the results, with five sweeps at five different bandwidth settings, from 1MHz to 63MHz (the maximum setting).
As you can see, the gain is flatter than 0.5dB overall, and for individual settings it’s better than 0.1dB. And now here’s a screenshot zoomed all the way out, with the bandwidth set at 50MHz. To achieve an ENOB of 8 bits, we need at least 50dB of rejection at our Nyquist frequency (125MHz) in order to prevent aliasing. Here we get 80dB of rejection, well beyond the requirement.
The frontend was working so well, in fact, that I decided to test it at 63MHz (the maximum frequency of our VGA Filter) and we were able to get 65dB of rejection! Therefore, we can comfortably rate the Probe-Scope’s maximum bandwidth as 63MHz, surpassing our design goal of 50MHz.
You can find the schematics and layout for this board on our GitHub, or if you go back to an older project log, there’s a screenshot of it too. Next on the list is the Digital Test Board, which we’ll have a short post about just verifying its hardware up soon.
We just got in the parts and PCBs for all our test boards, so this will be the first of several posts on the testing and bring up of the boards.
The first board we built and tested is the Frontend Divider test board (for a detailed explanation of why this board is so complex and needs careful attention, see our previous post). Below is the assembled Frontend Divider board. It contains a 30x resistor divider with special layout considerations, a tuning capacitor to ensure a flat response, and the frontend buffer amplifier.
Ideally, we’ll be able to tune the divider so that the frequency response is very flat; in order to test that, I connected it up to our Bode 100 Vector Network Analyzer, a low-cost, (relatively) low bandwidth USB VNA. It can perform several measurements over a range of frequencies (from 1Hz to 50MHz), most importantly for us, it can measure Gain Magnitude, Gain Phase and Group Delay.
Here is a picture of the Frontend Divider board hooked up to the Bode 100… do ignore the mess.
After a bit of fiddling with the tuning capacitor’s value, I was greeted by this beautiful bode plot:
The gain is almost perfectly flat across the board, and while the constantly decreasing phase might instill some concern at first, that too is our desired outcome. (It’s what we want in order to get a constant group delay, which we’ll cover in detail in a future article.)
Here’s some more detailed plots of the Gain, Phase and Group Delay (shown as Gain Tg):
Now, in order to measure past 50MHz and fully understand the performance of our Frontend Divider, we need to switch to another instrument: the DG8SAQ VNA from SDR-Kits. The DG8SAQ is a low-cost VNA which can measure from 1kHz to 1.3GHz. I swept our fronted from 100kHz to 250MHz. You can see from the plots below that the Frontend Divider is not the bottleneck of the Probe-Scope; it actually has a bandwidth of more than 150MHZ!
For the curious readers, the Frontend Divider schematic is shown below, and as with everything we do, the source can be found on our Github.
The frontend (input) divider of an oscilloscope is one of the most critical components of the whole system. It needs to attenuate the incoming signal a precise amount over a very wide frequency range; in the case of the Probe-Scope, from 1 Hz to 50 MHz or more. This requires careful consideration of the parasitics on the PCB.
First, it’s important to know what an ideal frontend looks like. It’s a resistive divider with a buffer, with a constant divide ratio vs. frequency:
However, in the real world, things are not as simple as they might seem. Every component has what are called “parasitics”, i.e. parasitic resistance, capacitance, and/or inductance: these are the result of unwanted interactions between elements. In the case of our divider, parasitic capacitance has the largest effect on the system. The first major contributor is the capacitance of the op-amp’s non-inverting input. We are using the Analog Devices AD8055 for our buffer, which has a typical Input Capacitance of 2pF. The effect of this capacitance on the circuit can be drawn in the schematic:
The next elements that contribute capacitance are the resistors themselves. The pads on the PCB that they sit on form parasitic plate capacitors with the ground plane below:
The formula for a plate capacitor is C=(k ϵ0 A)/d. Because the size of each pad is 700μm by 800μm, assuming a standard PCB thickness of 1.6mm, as well as a dielectric constant of 4.2, that gives each pad around ((4.2)(ϵ0)(0.56mm^2))/1.6mm=13fF of capacitance. Here’s what those look like in the schematic:
If we analyze the frequency response of that circuit, it’s clear that it won’t work for our target bandwidth of 50MHz. The parasitic capacitors form a low pass filter with the resistors, giving the front end a response that’s far from ideal:
So, since that doesn’t get the system where it needs to be, we need to compensate for it somehow. If we simplify the system, we see that there are two critical values, the capacitance across R1 and across R2 (where CR1 includes other parasitics, putting its value is in the 0.1pF range):
For the divider to have the same ratio for all, frequencies C_R1 needs to be one tenth the value of C_R2, which means it needs to about 0.5pF. Because the parasitics are not well controlled we need to add a tuning cap, but there’s a problem: tuning caps in the 0.5pF range are VERY expensive (in the $20+ range), while inexpensive turning caps start in the 2pF range. To accommodate an affordable tuning cap, CR2 would need to be in the 20pF range, however that would result in the whole frontend having a high capacitance.
Instead, we can intentionally add some parasitic capacitance to R1 and add a 2pF tuning capacitor to R2. This will give us the desired response and a small input capacitance, without driving up the cost. We achieve this by extending the node between R1 and R2 as a plane on the bottom of the PCB, and placing another plane above it that’s connected to the top of R1 (Vin). Below is a screenshot of the layout of the Probe-Scope input divider, where R1, R2, and R3 form the top resistor (R1 in the schematic, it ends up being a weird value hence the multiple standard resistors) and R4 forms the bottom (R2 in the schematic). The area circled in pink is the top plate of the parasitic capacitor, and the green arrows indicate the vias that connect the divider node to the bottom plate of the capacitor.
Here’s what our model looks like now:
And with all that in place, here’s our ideal frequency response!
We elected to put the frontend divider on its own PCB for the first round of prototypes, rather than integrating it with the rest of the frontend so we can test each component independently.
Here’s a picture of the board, it's slightly more complex than the example presented above. A detailed post about the frontend performance will come shortly!
In a similar fashion to the Analog Input Test Board, we created a Digital Test Board, so that we could compartmentalize development efforts and get both halves of our system up and running at the same time. The Digital Test Board contains the Analog Devices AD9481 ADC and everything needed to get data out of it and on to a computer over USB: to accomplish this it uses a Lattice MachXO2-4000 FPGA, a Cypress S70KS1281 self-refreshing DRAM, and a Microchip PIC32MZ2048EFH MIPS microcontroller.
The ADC is clocked by a Microchip DSC1103 250MHz LVDS MEMS oscillator, and since its output is 250Msps Double Data Rate (DDR), its output clock is 125MHz. It has two output ports, A and B, each 8 bits, with A being valid on the falling edge of the clock and B on the rising edge. The data clock output from the ADC is used as the clock source for the FPGA, which has built-in DDR support on the bottom IO cells.
Once the data gets into the FPGA, it’s stored into “EBR” SRAM in FIFO mode (a hardened feature of the MachXO2), where it waits until it can get send out in blocks to the Cypress “HyperRAM” device. Also at this time, the FPGA checks the incoming data for a trigger condition. We use the FPGA’s PLL to generate a 166MHz clock for the HyperRAM, which is its top speed. Since the HyperRAM is a self-refreshing DRAM, it does have to spend some time refreshing occasionally, and we also have to reset the address when we get to the end of the memory, so we can’t just write continuously at the ADC’s output frequency. Instead we run it slightly faster and write out bursts of data, allowing for control and refreshing to occur in between bursts.
The PIC controls the FPGA via SPI (using the FPGA’s hardened SPI peripheral) and also with simple GPIO. It sets the trigger conditions for the FPGA, and when a trigger is detected, it requests the relevant slice of data from the FPGA, which will transfer it out of the HyperRAM and into the PIC via a parallel port. At the same time (since in most circumstances the PIC cannot hold all the data at once), the PIC transfers the sample data over USB to a Python application running on the user’s computer. The design data for the digital test board can be found on our GitHub, and here is a direct link to the PDF:
Our boards have finally been shipped, so we'll get them hopefully on Wednesday and build them this week.
While designing our Digital Test PCB (which I'll be posting about in detail in a day or two), I decided to take a couple screenshots of the board as I worked on it, thinking it might be fun to see how it comes along. In the center of the board is the MachXO2-4000 FPGA, with our PIC32MZ2048EFH processor on its left port, the AD9481 ADC on the bottom, and a 128Mbit Cypress HyperRAM on the right port.
I start by defining the board size and bringing in all the components from the schematic, then placing the connectors and major components:
The board is 100x100mm as that's the cutoff for PCBWay's lowest price. I organized everything around the FPGA since it's really the hub of all our signals. You might notice that the final version has a few more connectors - since we're in the early development phases, it's a fairly fluid process, and Mark thought of some more nice-to-have features so I just went ahead and threw them on.
Next, I get all the decoupling capacitors next to their associated ICs (with some old-fashioned pen-and-paper action) and roughly place minor components:
Now for the "draw the rest of the owl" step, I route everything and finalize component placement. This board took me roughly 10 hours to route (very roughly, I should have timed it). I put a ground plane on layer 1 and a power plane on the bottom, so I tried to avoid those as much as possible but completely avoiding them wasn't practical. I'd have gone into a bit more detail with the pictures, but it's really just drawing lines that can't cross for hours until it's done. Here's the result:
Of course, there's always still a few tweaks to be made. I made the power test points a little more logical, moved our debug differential IO to the top port which has true differential drivers, and made a single trace going under the RAM a couple mils thinner as the manufacturer informed us that there is a wider clearance between BGA pads and other copper in the normal process (it would have cost around $400 instead of about $50!). We also realized just in the nick of time that we required the 1.8V version of the HyperRAM part in order to get the full speed which we needed, so a 1.8V rail was added for that.
So here's the final design we sent to the manufacturer:
On the final board, I'm planning to get a timelapse of the whole layout process (unfortunately none of this layout will be reusable as compactness is a priority). You can check out the board on our GitHub page, and I will be writing another post about its design and function in a couple days. Hopefully we'll have them built up a few days after that!
We created a test board for the analog input section of the Probe-Scope to test the frequency response and the noise floor. This board contains the input divider/buffer, vertical offset amplifier, and programmable gain amplifier (with filters); it connects via SMA to a companion digital test board which hosts the ADC and necessary acquisition hardware. The schematic of the analog test board is shown below.