I’ve been thinking about building stuff with FPGA’s for a while, and usually get turned away because FPGA’s are considerably harder to implement than microcontrollers since they have no on-chip memory. It is necessary to re-program the gates every time they power up, which requires an external flash memory chip. There aren’t great references online for the DIY community, so I figured I’d post how to get this working. Not using dev boards opens a world of opportunities, so I’m a proponent of not using Arduino’s and their FPGA equivalent for too long (sure, they’re good to get started with, but don’t become dependent)
Not wanting to screw up an expensive complex board by being a first-timer at putting an FPGA into a circuit, I figured I’d build a little test board with the cheapest Spartan 6 you can get (about $10), which comes in a solderable TQFP144 package. Sadly, most high end FPGA’s are BGA and therefore quite hard to solder as a DIY project.
There are quite a few ways to program an FPGA. The most common would be direct programming, indirect programming, and processor programming. Direct programming is where there is a PROM chip which is programmed over JTAG, and the PROM chip programs the FPGA at boot. Indirect programming is where there is a PROM chip attached to the FPGA, and JTAG also connects to the FPGA. In this scenario, JTAG can either directly program the FPGA, or load a temporary PROM-programmer application onto the FPGA which programs the PROM and then restarts. The operational differences are subtle, but the circuit is quite different. Processor programming is where a microcontroller on the board performs the task of programming the FPGA, which would require custom code in the microcontroller as well as requiring you to have a micro at all. While substantially more advanced, this is way more powerful because the microcontroller can dynamically reprogram the FPGA on the fly for different tasks.
For this project, I’m going to use indirect programmimng, because it’s about $8 cheaper than direct programming, and processor programming is vastly overcomplicated for what I want to do.
The first task is to select a PROM chip. There are a number of series that are recognizable by the XILINX iMPACT tool. The M25P series is cheapest and very popular. You’ll need to read the datasheets to figure out how much memory your FPGA needs. I’m using the Spartan LX9 in my production board (although LX4 on this test board since it’s half the price), so I selected a 4M PROM. The M25P40 is only 67 cents on digikey in qty 1!
I’m using the Digilent JTAGHS1 programmer, since it’s the cheapest programmer that’s guaranteed compatible with Xilinx iMPACT (the programming application provided in ISE, the Xilinx IDE). At $55, it’s not the cheapest thing but you only have to buy it once and it’s way cheaper than the Xilinx brand programmers.
Here is the minimum schematic to hook it up to a Spartan 6 for JTAG programming. Other micros will have slightly different configuration bits, but the concept will be the same:
P144, HSWAPEN, if low enables a pullup resistor on all pins during configuration so any tristate devices are defined. You’d probably want to do this.
When P69 M0 is high and P60 M1 is low, the chip is configured for indirect programming over JTAG. Other M-configurations would be for other programming modes such as direct programming. For other chips, you’ll just have to look up whatever configuration bits your chip uses.
P37, PROGRAM_B_2 is the main chip reset pin. It should be pulled up, and to reset pulled low.
P71 DONE is a “done programming” output indicator provided for convenience.
TMS, TDO, TDI, and TCK are the JTAG pins, which just get hooked up to your programmer.
CLK, MISO, MOSI, and SS are the SPI data lines to the PROM.
While this guide isn’t meant to be a “how to program for FPGA’s”, I’ll include my test script here, which just sets a few pins to test out the programming. The user constraint file tells the compiler what pins connect to what module IO lines:
Now, after generating the programming file, select configure target device from the process list and iMPACT will load. To just program the FPGA over JTAG without worrying about the PROM, select Boundary Scan, right click, select Initialize Chain, and if the FPGA is connected it should almost instantly detect it. You’ll then be prompted for a BIT file to program it, or you can right click the symbol for the chip and select a file that way. It’ll also prompt you to ask if you want to generate a PROM file. If you’re just loading the chip once, click no. Right click the symbol, hit program, and it’ll program the FPGA.
Now, if we want to make the program persistent on the FPGA, we need to program the PROM which is only slightly harder. This time, select yes to make a PROM programming file or alternatively select “Create PROM File (PROM File Format…)” from the iMPACT Flows list. This will start the PROM File Formatter:
In this window, we tell iMPACT exactly what to make as a PROM file. In this case, we’re using the M25P chip which is SPI flash (but not Xilinx Flash/PROM, which is for direct programming). Next select the storage size, in my case 4M. Finally, give it an output file name, and select OK. It should ask you for a BIT programming file, which is the programming file that you otherwise would have sent directly to the FPGA. It’ll then ask if you want to provide another BIT file for a second FPGA (you can program multiple FPGA’s off of a single PROM chip). If you just have one like me, select no.
An easy to miss step is to select Generate File from the iMPACT Processes list. When you do this, it’ll actually make the PROM programming file (I think it was .mcs? I forget). Now you can go back to Boundary Scan, right click the FPGA, and add the PROM file. A FLASH module should be appended to the diagram, which you can right click and Program:
That’s it! Now every time you turn on the board, it will auto-program off of the PROM. If you want to change the program, just repeat the process of uploading a new PROM file.
I haven’t blogged in 14 months! This is not acceptable! I’m sorry Charles!
Let’s see what I’ve been doing in the last 14 months…
I continued making better and better electrostatic headphones
The art of stretching 1 micron and 2 micron mylar film for electrostatic membranes:
MITERS got new drill bits and end mills:
I finished my DIY SR-007 headphone clones, version 1
Went home for January and climbed a bunch of mountains
Came back to Cambridge, and built my Mini-electrostats… they didn’t sound very good, I dismantled them and deemed it a poor design
Then there was a huge snow storm (just barely not a blizzard, technically) in Boston where we got 3 feet of snow
With only a week left in the calendar winter to check it off the list, rented a car and did a midnight solo of Mt. Washington
I finished my SR-007 clone DIY headphones, version 2
Then I went back to Washington and worked at Boeing for the summer designing pulsed power electronics stuff in the physics applications lab. On the weekends, I climbed lots of awesome peaks. Now have 3/5 of Washington’s volcanoes checked off the list (Baker, Adams, and Glacier Peak [the best peak in Washington])
And had a mountain goat wander by while I had my camera out
… so I used it to make DIY Sennheiser Orpheus clones, which sound spectacular. They are still a work in progress
Learned how to analog in 6.301
Experienced 100+MPH wind and -40F windchill in a fruitless solo endeavor on Franconia Ridge
Brought my DIY headphones and Blue Hawaii to New York for a head-fi meet
Ok I’m going to stop procrastinating and get back to studying. Only another week of classes and then finals!
I’ll be going home for winter break, and don’t want to carry my big electrostatic amplifier, so I decided to make a Class-D portable amplifier.
Here are a few schematics… Eagle is too much of a pain to make nice schematics, so I’m not going to post a full schematic at this point.
Aside from the actual amplification stage, I need to make 300V rails and and a 350V bias rail (configurable for up to 580V) to charge the diaphragm in the headphones. To do this I’m using the LTC3803 flyback controller. This is a current-mode control chip with an external compensation network in a convenient SOT-23-6 package. I wanted to test this out before ordering the full board, so I built a prototype converter for the 300V rails (if that works, the DC bias flyback will almost surely work).
Tests show it working fairly well, although it draws quite a bit of no-load power. I attached a BJT as the load so I could vary the output current to plot efficiency vs. output, which resulted in the following data:
I fit the equation
which represents a constant controller current (including core losses with the output load of the feedback resistors). This fit results in a controller power draw of 1.25 0.05W and an efficiency of 0.78 0.04. This is a bit high on the controller power draw, which I may be able to get a bit lower by not using the shunt regulator in the controller. Efficiency could be better, but is quite tolerable given that the amplifier really won’t draw much of any current on the 300V rails.
With summer comes another few weeks at home to spend time in the mountains (and in the snow since it was a pretty heavy winter for snow so lots still around).
All trips were solo, just the way I like it!
First up, Mt. Daniel (7/21/2012):
A failed summit but an awesome trip. Was feeling sick 1000ft below the summit, so I turned around. Still a great place, definitely will return sometime and probably make it a two-day climb.
Next, a quick run to Kendall Katwalk (7/24/2012):
Quite a bit of snow left for so late in the season. A number of very steep snow crossings before getting to the Katwalk, although there were good boot paths through everything. After the Katwalk, almost all snow covered.
Last, Mt. Adams before heading home to Boston (7/29/2012):
Did this one as a 2-day trip, and made it to the summit. Of course, the one thing I somehow forgot to bring was my camera since it was on the charger overnight. Oh well, iPhone will do!
Summer is almost over with classes starting up again next week.
I completed my Blue Hawaii amplifier over the summer, made a number of prototype electrostatic headphones (working on a final version now), and built a tube phono stage.
I ended up having some thermal issues on hot days after a long on-time. I recently put a bunch of holes into the bottom plate and in empty space on the PCB to hopefully allow for convection to occur more. Hopefully that’ll help, won’t have a chance to test it until I get the new headphones finished.
As for the headphones, there was a lot of trial and error learning done there. I’m sure I won’t remember even half of what I had to figure out, but here are some details on that part of the project:
Electrostatic headphones operate by moving a thin charged film between two high voltage differentially driven stators. The distance between the stators is 1mm in my drivers, many retail headphones go even lower. This film has fewer resonances than a conventional driver, but it has one pretty severe resonance in the range of 100-300hz, since after all it is basically a spring and mass system. This resonance can be damped out by having a good seal between the driver and the listener’s ear. In fact, this drives the resonant frequency lower by creating that volume, but more importantly damps it. Below the resonance, however, there will be some bass roll-off if the resonance is too high. This means the tension of the film (in this case, I’ve been using both 2uM and 6uM mylar, will probably stick with 2uM) has to be low enough to keep that resonance low. But, the tension must be high enough to give it enough spring to not just stick to one of the stators. Since there’s only 0.5mm between the film and either stator, this provides the opportunity for the film to pretty easily stick if there isn’t enough tension to pull it back to the center.
The stators I made are 0.062″ FR4, which I manually drilled with a small drill-press using a perforated steel sheet as a drill template to ensure a clean pattern:
Through lots of trial and error, I found a good tension just above the point of instability where the film hit a stator and stuck. To accomplish the tensioning, I hang weights from the film as seen in this picture:
Frames are then glued down onto the film to hold the tension.
Next, these films must be coated in a partially conductive material to allow the placement of a charge on them. There are quite a few methods people use to accomplish this. I started out using a white-glue/graphite/water mixture which did work, but is somewhat hydrophilic which would not lead to long term reliability. I ended up using an Elvamide solution, which leaves a coating of nylon behind after drying which is just conductive enough to carry the charge out onto the film.
To accomplish this coating I have just been brushing it on, but in my final revision I am going to airbrush it on since I’ve had trouble getting a uniform coating with a brush. When the coating is not uniform and equal on both sides, it can lead to channel imbalances.
A mylar film glued to the frame and coated with Elvamide:
Next in the process was figuring out how to make a protection film for the inside. Since sweat can build up while listening, this could lead to moisture getting into the driver if run with no protective film. Initially I just put another tensioned mylar between the driver and the ear, but found that to cause bass dropoff (green without protective film, red with):
Having even enough tension to make the protective film structurally stable caused some amount of bass roll-off.
I noticed almost all retail electrostatic headphone drivers use a wrinkled material as the sweat protection screen, and figured I’d give that a shot. This seems to greatly lower the material tension, but allows it to still stay structurally rigid.
Over the last week or two, I’ve been learning solidworks to design some more reliable headphones. In this revision I will be making all parts by CNC (with the exception of the stators which have been drilled with more precision using a complete drill mask for all holes including mounting holes)
I’ve been working on making a Blue Hawaii electrostatic amplifier (designed by Kevin Gilmore). I’ve spent a lot of work on the chassis which I built myself. I’ve been having some trouble with premature clipping in the amplifier which I still need to sort out, and also need to figure out why there seems to be bandwidth issues above 10khz. Other than that, the project is well on its way to completion.
To do toner transfers for double sided boards, I hold the papers up to a light and align them, then tape the papers together to form a sandwich, and slide the board in before transferring. I’ve never had it off by more than a quarter of a drill hole. Usually it’s practically dead on.
More to come as I put more work into the project.
I had a fairly crude monophonic interrupter built for big-coil operation, but I wanted to improve on its design. It had no software control over pulse width so every note had the same width, and of course polyphonics is more fun.
One critical design requirement was that I need hardware protection against the controller accidentally going CW, even if the microprocessor somehow screws up. This is relatively easy by using a 555 as a one-shot on each rising edge, then AND gating that to the actual signal so that if an abnormally high duty cycle (or more likely a duty cycle of 100%) is attempted, the one-shot will turn off at a pre-set time (determined by trimpot R4). It is worth pointing out that an inverter could be used to trigger the 555, which requires a downward-going edge to trigger. I chose to use a second microprocessor output since it saves on components (and I wanted as many gates buffering the output power as possible).
I also wanted to have a fixed oscillator in the interrupter so that I can still use it as an easy optical oscillator for bench testing coil controllers without the need to setup a MIDI device. I threw in another 555 for this. A SPDT switch on SV1 allows for selecting between MIDI or this fixed oscillator.
Another goal was to have the controller programmable over the MIDI interface to control parameters such as pulse lengths specific to note (allowing low notes to be on for a longer bang than higher notes), MIDI channel, and multipliers to control pulse width during polyphonic playback. This can all be programmed over SYSEX, a communication means within the MIDI protocol that is designed for flashing MIDI hardware with firmware or updates.
I put together a quick python script to allow for changing parameters, although I may make a GUI at some point.
This project was also the first time I’ve tried 0.4mm isolation on ground planes, which allows for ground planes to pass between pins of a DIP socket. It etched fine using the laminator for toner transfer. The etch precision is getting so good that I need to look into solder masks.
I setup a laminator at MITERS for etching a few months ago, but forgot to write about it until now. So here’s some pictures and a little info on the modifications to make it into a fantastic PCB etcher:
It is a GBC 9″ laminator (pretty cheap, $45 on amazon usually, I’ve heard you can get it for $10 or even less if you get lucky on sales or search around a lot). Out of the box, it’s an incredibly low quality laminator that can’t do much more than paper. If you put a circuit board through it, the gears in the motor strip themselves out and it quickly destroys itself. You can kind of force it to keep going by pushing the board in, but it’s totally not worth it after how great it performs with some modifications.
All I did was take it out of its case and put a new motor on it. I put a slower motor on, something like 2RPM instead of the stock 6RPM motor (I forget the exact numbers). You can’t really go too slow since you want lots of heat and pressure to get a circuit board to transfer well, and multiple passes aren’t ideal since it can lead to blurring if thin paper is used.
I threw together some mount brackets which weren’t quite strong enough to prevent the motor drive gear from skipping, so I made a little bearing to keep the motor shaft lined up with the gears. That prevents the gears from slipping away from each other, and it easily feeds standard PCB thicknesses.
The capacitor on the front is a motor start cap which is overkill in size for this motor, but it was laying around so why not use it.
Transfers come out 100%, every time. There’s no longer any guess work involved with the time and pressure of using a clothes iron. It makes it a real treat to etch boards.
I’m currently bypassing the slow reverse recovery diodes in the IGBT bricks with much faster minibrick diodes.
I am adding a diode to the IGBT path to turn it into a switch that blocks both directions, and conducts in one direction when on, then adding a fast diode across that assembly.
The blocking diode doesn’t need to have a voltage standoff rating of the full bridge, because it is in series with the IGBT brick which can stand off the full voltage. Ideally I would use the highest current and fastest diode available, with no regard to standoff voltage. Since a friend had some laying around, I used some DSEI2x101 minibricks for this.
For the bypass diode, full voltage standoff is required. For this position, I’m using ST 12012TV1 diodes.
I’m a bit concerned about passing all this current through these diodes, but it is well within the pulse and I^2S rating of the components so hopefully it will be fine. All the minibricks will be heatsinked to busplates.
Some shiny pictures of the modifications in progress:
I’ve been pretty busy with class work so I haven’t been building a whole lot recently. I did get a polyphonic controller working for hatcoil, though. By using both timers on the Atmel based controller, it can simultaneously play any two notes making the songs much more interesting to listen to.
Also did a bit of dry ice overclocking, although the board seems to be damaged and won’t get basically any overclock stable (used to run 3.6ghz stable on water and can’t even get it up to 3ghz anymore)