|
One more note - Here is the design flow of FPGA programming, at least for me. 1) Work out the rough details of a design on paper, or maybe in my head. 2) Write them into Verilog code (I usually work in Xilinx ISE, which comes with a pretty good hilighting text editor) 3) Simulate the Verilog code (Xilinx ISE comes with a free simulator that works pretty well) 4) What the gently caress I made a typo or something, go back to stage 2. 5) After a few repetitions of this my design is "complete" 6) Synthesize the design, which is basically the first stage of compiling your code into a binary file for your FPGA. 7) If I wrote lovely code and it didn't synthesize, I have to gently caress around with it until the synthesizer will accept it. 8) Once the design synthesizes, simulate it one more time for good luck. 9) Run the the "implementation" tools, which takes your synthesized design and turn it into a binary file you can load onto your FPGA. These tools include the translator, the mapper, the placer, and the router. They're usually all just one big step called "implementation" though, most people don't mess around with doing them one at a time. 10) Drink some beers, you're done!
|
#
?
Mar 11, 2009 18:28
|
|
|
|
| # ? May 21, 2024 22:18 |
|
|
I'm having trouble understanding the purpose of FPGAs, and to what sort of problems they are applied. With a microcontroller, the hardware stuff is already taken care of and you just have to focus on programming the thing. If a microcontroller is lacking in functionality, why not just use an embedded processor or a PC? I apologize for the naivety of this question.
|
|
#
?
Mar 12, 2009 06:37
|
|
|
Cyril Sneer posted:I'm having trouble understanding the purpose of FPGAs, and to what sort of problems they are applied. Cost or performance issues usually. With FPGA you can design the exact logical structures you need for an application. So if I can use a 1.50$ FPGA rather than a 5$ DSP I'm ahead. Or I could use them to interface certain devices to one another such as ram to a cabled PCIE interface. Glue logic basically. Or I could devote one part of the FPGA to application specific logic and the rest to operate as an ARM mcu. There are not always chips that interface to everything you want to hook up to so thats why fpgas are popular.
|
|
#
?
Mar 12, 2009 07:28
|
|
|
One FPGA is worth a hell of a lot of 74-series ICs.
|
|
#
?
Mar 12, 2009 07:29
|
|
|
Cyril Sneer posted:I'm having trouble understanding the purpose of FPGAs, and to what sort of problems they are applied. Custom logic. DSP. Highly defined timing characteristics. Most application where you would have previously made an ASIC. All in the same device with the ability to reconfigure it to do something different. With a reasonably sized FPGA I can do a streaming 4K FFT at 100 MHz, with room left over. That amounts to 25 thousand, 4K FFTs per second. Try that with a microcontroller or embedded processor. If all you need is a microcontroller, then yeah an FPGA probably isn't the right device. But lets say you need your micro to talk to a compact flash or ATA device at full rate. Now the ability to create custom logic is a lot more attractive. And you can still embed a microcontroller into the FPGA, provided it's big enough.
|
|
#
?
Mar 12, 2009 07:47
|
|
|
Cyril Sneer posted:I'm having trouble understanding the purpose of FPGAs, and to what sort of problems they are applied. An FPGA has a lot of I/O. I mean a lot. A large 8-bit microcontroller has what, maybe 4 8-bit I/O ports? 5 tops? That is 40 possible I/O pins, each port of which has to be controlled one-at-a-time. Writing a 40-bit value from RAM to pins would take probably 10-12 clock cycles on a microcontroller. On an FPGA, you could write a 300-bit value to I/O headers all at once in a single clock cycle. Or say you were doing some signal processing that required 50 operations to happen in parallel (such as an FIR audio filter). On an FPGA, you can just do them in parallel. On a microcontroller, you have to figure out some way to cram them into the number of available clock cycles. Let's say you have some kind of custom interface that uses 12 7400-series logic chips. Whatever that interface does, you could easily do using a single CPLD. This takes less board space, it's cheaper to buy parts for, and would be more reliable because there are fewer failure points. And while a CPLD could replace a couple dozen 74-series chips, a small FPGA could easily replace hundreds of 7400-series chips. A large FPGA could do jobs that it would take thousands of discrete chips to implement. So basically, FPGAs are like an ASIC (a custom integrated circuit), except you don't actually have to go out and get an ASIC fabricated - you can just use an FPGA instead and reconfigure it as many times as you like. Let me go into detail on a small FIR filter (for example) The microcontroller assembly code for a 3-tap filter would be something like: SET SUM = 0 LOAD NEW VALUE LOAD OLD VALUE 1 FROM RAM MULTIPLY ADD TO SUM LOAD OLD VALUE 2 FROM RAM MULTIPLY ADD TO SUM LOAD OLD VALUE 3 FROM RAM MULTIPLY ADD TO SUM. STORE SUM IN RAM INCREMENT RAM POINTER OUTPUT SUM. Even if your microcontroller has a built in multiplier, you're still looking at 15-20 instructions or so per sample. If it doesn't have a multiplier, you are looking at a couple hundred instructions. An FPGA could perform the entire operation in one single clock cycle because you could just make your design have three dedicated multipliers, three dedicated registers, and three dedicated adders all connected to each other. Poopernickel fucked around with this message at 08:44 on Mar 12, 2009 |
|
#
?
Mar 12, 2009 08:25
|
|
|
Cyril Sneer posted:I'm having trouble understanding the purpose of FPGAs, and to what sort of problems they are applied. One of my current projects is the design of a board that is, in part, the graphics controller for an LCD monitor. An FPGA is being used as the main graphics controller because it allows us to update the product or make changes without having to totally redo the board (and all the expensive and time consuming compliance testing required for new boards). This came about due to the volatility of the graphics controller market. We've done 3 or 4 iterations of this board because it seems like graphics controllers end-of-life every 6 months and we have to design in a new one. Using an FPGA solves this problem and gives us control of the life of the main component on the board. I also designed an FGPA board when we were creating a new ASIC so that we could test most of the ASIC before committing to the hundreds of thousands of dollars it costs to make an ASIC. In this case the FPGA board was around $3-4k to make, but it caught a lot of things that would have meant very expensive re-spins to the ASIC had they made it through. It also allowed our firmware team to begin coding for the ASIC before they had actual hardware to work on.
|
|
#
?
Mar 12, 2009 13:31
|
|
|
Hillridge posted:I also designed an FGPA board when we were creating a new ASIC so that we could test most of the ASIC before committing to the hundreds of thousands of dollars it costs to make an ASIC. In this case the FPGA board was around $3-4k to make, but it caught a lot of things that would have meant very expensive re-spins to the ASIC had they made it through. It also allowed our firmware team to begin coding for the ASIC before they had actual hardware to work on. Were they able to use the same logic/HDL for the FPGA for the ASIC without much modification (I/O excepted)? I don't know anything about how ASIC designs are specified in practise. For my final year project, I was originally planning to use a DSP, probably an older TI one since that's what the department mostly supports, but now it may be looking like a microcontroller is just as good a solution. Perhaps a Cortex M3, though a DSPic has been mentioned as a viable candidate. DSP: -Faster for MAC operations, ie. FIR filtering -External memory, program code would have to be on an external PROM, loaded onto external SRAM for running -Have to learn a different language Microcontroller: -Higher clock speed (at least when compared with older DSPs) -Internal memory, less loving around -Easier and faster to upload to -More modern development tools I/O requirements are: 3 ADCs, sampling transducers of power line frequency + harmonics, so pretty low bandwidth, multiplexing possible 6 GPIO for a full bridge inverter gate drive
|
|
#
?
Mar 12, 2009 21:33
|
|
|
Another way to program an FPGA is use simulink (matlab) with the xilinx blockset and have that automatically generate your code.
|
|
#
?
Mar 13, 2009 13:29
|
|
|
Bush Ant posted:Another way to program an FPGA is use simulink (matlab) with the xilinx blockset and have that automatically generate your code. Otherwise known as cheating. 
|
|
#
?
Mar 13, 2009 14:50
|
|
|
catbread.jpg posted:Were they able to use the same logic/HDL for the FPGA for the ASIC without much modification (I/O excepted)? I don't know anything about how ASIC designs are specified in practise. Hillridge fucked around with this message at 17:37 on Mar 13, 2009 |
|
#
?
Mar 13, 2009 17:32
|
|
|
Bush Ant posted:Another way to program an FPGA is use simulink (matlab) with the xilinx blockset and have that automatically generate your code. There's also labview FPGA, but you have to use an NI board. Several companies also make tools to turn C code into FPGA logic, but I've never used them. They tend to be very expensive and it always seemed like way too much could go wrong.
|
|
#
?
Mar 13, 2009 19:13
|
|
|
catbread.jpg posted:DSP: I wouldn't worry about the absolute clock speed at all. It is almost meaningless when comparing across architectures, such as a DSP to a microcontroller. Just about the only way to tell which is better would be to benchmark both and see where they fall. If you can put your algorithm into a microcontroller and have it still run fast enough, just stick with what you know. Do not undervalue to the benefit of "less loving around."
|
|
#
?
Mar 13, 2009 19:20
|
|
|
SnoPuppy posted:There's also labview FPGA, but you have to use an NI board. LabView is a hell of a lot better suited than C for an FPGA. How exactly would that work? It would be like putting everything in a procedural block, completely negating most of the uses of an FPGA, wouldn't it?
|
|
#
?
Mar 13, 2009 19:52
|
|
|
hobbesmaster posted:LabView is a hell of a lot better suited than C for an FPGA. How exactly would that work? It would be like putting everything in a procedural block, completely negating most of the uses of an FPGA, wouldn't it? You're right about labview - its pretty obvious to see how stuff can run in parallel and where you could add pipeline stages. As far as C goes, I don't know exactly how it works, but I think the compiler is able to identify instructions which can run in parallel. Like if you had a function to compute a dot product, you could do all the multiplies at the same time and then follow it with an add. I'm sure its *heavily* dependent on the way you write your code.
|
|
#
?
Mar 13, 2009 20:30
|
|
|
hobbesmaster posted:Otherwise known as cheating. N  Seriously though the MATLAB approach costs some serious mega bucks. A MATLAB/Simulink/toolboxes yearly license is around $6000, and the xilinx tools to do that will run you another $2000 or so. This is not a great approach unless you are big pimpin' even  can't save youPoopernickel fucked around with this message at 00:38 on Mar 14, 2009 |
|
#
?
Mar 14, 2009 00:28
|
|
|
Cyril Sneer posted:I'm having trouble understanding the purpose of FPGAs, and to what sort of problems they are applied. For all practical intents and purposes, it is a poo poo-ton of nand-gates...AKA universal gates...AKA you can make any circuit you want on them without having to buy a random IC gates... This means that you can make it do whatever any other micro-controller, microprocessor, or other IC does, along with whatever random circuitry you want it to do...at the same time, without having to put up the cash to dope custom silicon.
|
|
#
?
Mar 14, 2009 02:26
|
|
|
Nerobro posted:A friend on freenode told me I needed to buy this: http://www.seeedstudio.com/depot/digital-storage-oscilloscope-with-panels-p-167.html It's a aurdino based 1mhz Digital Ociliscope. It arrived in the mail two days ago. edit: Nevermind. I guess it was a temporary bug or something. armchair fucked around with this message at 03:02 on Mar 14, 2009 |
|
#
?
Mar 14, 2009 02:57
|
|
|
armchair posted:So I was interested in this and took a look today to buy it and it looks like they have yanked it from their website. I was able to see it back on Wednesday. It's pretty weird that it just disappeared from their website. I have one of these. Beware the .1 puff caps...I turned one from a half-grain of rice into a quarter grain of rice with one moment of over application near the bottom-right-corner of U4...that's with a microscope...Beware this thing; Caveats aside, it's a fun project if you can avoid Caffeine for a day.
|
|
#
?
Mar 14, 2009 03:28
|
|
|
Uh. Yeah. This is rather tricky. I wish i had some solder paste and a hot air pencil. I probally have it about half done. Do yourself a favor. Pay the $16 to get the assembled version. My time is worth more than $14 an hour, and I've put 90 mins in it so far.
|
|
#
?
Mar 14, 2009 08:09
|
|
|
my university requires engineering majors take an intro to programming course, and when I took it all we learned was Java (pretty much useless for engineers). Apparently this semester they changed it so that it's broken into sections, some of which focus on labview, matlab, and VHDL. I can't believe I missed out on this stuff. My inability to use development environments like these is going to burn me later... Still can't wait to get into FPGAs next semester though. I've found several places where they would work much better than microcontrollers. Mainly for precise timing operation, or for dsp. Does anyone have any experience with ATMEL's FPSLIC devices? They're essentially a hybrid of a avr microcontroller with a FPGA (up to 40K gates). At first I assumed they would be crippled and gimmicky, but I had the teacher of the FPGA class look at their documentation and he seemed impressed. On one hand I know that it's kind of redundant to merge a uC core with a PLD, since you could get pretty much the same thing with just a PLD by programming the uC core in, but I was thinking this would be a good route for me since I'm already comfortable with a lot of atmel's stuff, thus making debugging the logic stuff easier. here's a link to the FPSLIC page Also, what's the deal with configuration devices? It strikes me as ridiculous that FPGAs need them, and they're sometimes drat expensive (for our sodar project we had a $40 cyclone III and the config device cost $16). One of our group members actually mused about the idea of getting a simple microcontroller with a large memory (microchip makes some) and program it to act as a configuration device. Actually seemed like a really good idea, and I'd likely develop it if I had the time or expertise. But even then, why should the configuration memory be stored in a separate device? Would it really be so hard to put some flash on the FPGA itself? ANIME AKBAR fucked around with this message at 15:15 on Mar 15, 2009 |
|
#
?
Mar 15, 2009 15:09
|
|
|
ANIME AKBAR posted:my university requires engineering majors take an intro to programming course, and when I took it all we learned was Java (pretty much useless for engineers). Apparently this semester they changed it so that it's broken into sections, some of which focus on labview, matlab, and VHDL. I can't believe I missed out on this stuff. My inability to use development environments like these is going to burn me later... I'm not so sure that a labview, matlab and VHDL class would be a good introduction to programming. Though, I decided from the beginning that I would get a minor in CS and didn't touch MATLAB until circuits II which I took after completing the basic three semester programming sequence (all in C++, third class introduced PERL and a bit of Java too). So long as you are familiar with how a programming language works, any of them shouldn't take too long to figure out. Labview is interesting. Its very easy to slap something together that reads data off an instrument but it can get very difficult very quickly to have a larger application that actually works. The other problem with LabVIEW is that you are at the mercy of National Instruments for everything to keep working, they're not fans of backwards compatibility.
|
|
#
?
Mar 15, 2009 18:17
|
|
|
BattleMaster posted:One FPGA is worth a hell of a lot of 74-series ICs. Ya, but wouldn't it be fun to wirewrap all those 74-series TTL chips?
|
|
#
?
Mar 15, 2009 18:22
|
|
|
Jailbrekr posted:Ya, but wouldn't it be fun to wirewrap all those 74-series TTL chips? Especially if you have a run of 2000 units or so.
|
|
#
?
Mar 15, 2009 18:25
|
|
|
Jailbrekr posted:Ya, but wouldn't it be fun to wirewrap all those 74-series TTL chips?  I'm considering buying an FPGA development board because working with a Z80 with I/O and memory management logic built in 74-series is pissing me off. With an FPGA I could put all that crap on one IC with external RAM and ROM. Those Atmel FPSLICs look badass, too. The main advantage is that RAM and ROM with FPGA gates is kind of wasteful (right?). Not to mention that there may not be an AVR core available if you wanted to build your own.
|
|
#
?
Mar 15, 2009 18:41
|
|
|
Hey guys what do you think of this audio amplifier circuit? It's my first complete amplifier circuit ever, it does pretty much what I want it to do (according to the simulator). It's designed to deliver 100W into 8 ohms. I haven't designed the power supply yet but what do you think? Constructive criticism only please.
|
|
#
?
Mar 16, 2009 00:07
|
|
|
Oleum Animale 68C posted:Hey guys what do you think of this audio amplifier circuit? Looks neat! I'm a big fan of FET-based amplifiers.  It's been awhile since I took transistor design theory, but off the top of my head I would say take a look at the power dissipation in your 1 Ohm and 10 Ohm bias resistors. It is probably much higher than you think. Also how are you planning to generate the rail voltages?
|
|
#
?
Mar 16, 2009 02:58
|
|
|
Those FETs are gonna be eating up an awful lot of power, especially near the 1 and 10 ohm resistors. Also, even at lower voltages there's a significant blowthrough because at Vin=0V, your transistors are all partially on, and letting as much power through as they can. It's an interesting design, I'm not sure what all needs to be done to make it workable. It looks a lot like a 3 stage half-hbridge design, which is very intriguing. You have the makings of a class D amplifier, although something tells me that's not what you were going for.
|
|
#
?
Mar 16, 2009 07:00
|
|
|
clredwolf posted:Those FETs are gonna be eating up an awful lot of power, especially near the 1 and 10 ohm resistors. Also, even at lower voltages there's a significant blowthrough because at Vin=0V, your transistors are all partially on, and letting as much power through as they can. Yeah - I think it looks more like 3 class B amps ganged together though. Or 3 CMOS inverters. All pretty much the same thing though. And you're right about the power dissipation in those resistors. In the 1 ohm case, you burn V^2 watts of power. So for even 10 volts you have to dissipate 100 watts.
|
|
#
?
Mar 16, 2009 07:13
|
|
|
ANIME AKBAR posted:my university requires engineering majors take an intro to programming course, and when I took it all we learned was Java (pretty much useless for engineers). Apparently this semester they changed it so that it's broken into sections, some of which focus on labview, matlab, and VHDL. I can't believe I missed out on this stuff. My inability to use development environments like these is going to burn me later... Please tell me your engineering course is mechanical engineering or civil engineering I was taught html, java, c, c++, assembly, labview, matlab and simulink, ewb and pscad. We use atmel microcontrollers with their devopment boards and xilinx dev boards and a tonne of other specialized software and hardware, even the CRO's interface with matlab and capture software on the PC's. Infact I dont think I have been in a class in the last 3 years that hasnt done something along the lines of programming, we use matlab as a general purpose calculator as well, for solving matrix equations etc.. and getting plots of all sorts of things. We even use opnet modeler to simulate various computer network topologies and protocols
|
|
#
?
Mar 16, 2009 10:00
|
|
|
Oleum Animale 68C posted:Hey guys what do you think of this audio amplifier circuit? okay, in all honesty you need to take a step back and do some more research on how amplifiers work. I can guarantee this circuit won't perform the way the simulator indicates, since the falstad simulator doesn't take a lot of things into account. It assumes that FETs will be operated under certain conditions, which you're not. I can't tell whether you were going for class A or class B stages. Your N channel fets are common drain (like B class) and your P channels are common source (like class A). You don't want to mix and match like that. Also, I don't see any reason for those 1 and 10 ohm resistors to be that low. Making them 10K or 100K would give more gain (since your fets wouldn't be saturating nearly as much), not to mention saving a ton of power. Overall, I hate saying stuff like this, but it's more worth your time to scrap this circuit and get a better idea of how amps work than press forward with this design. It's completely infeasible. Also, aren't most 8 ohm loads only good up to 40 or so watts? I thought stuff around 100W was more like 4 ohms. Bush Ant posted:Please tell me your engineering course is mechanical engineering or civil engineering ANIME AKBAR fucked around with this message at 14:15 on Mar 16, 2009 |
|
#
?
Mar 16, 2009 14:11
|
|
|
ANIME AKBAR posted:But even then, why should the configuration memory be stored in a separate device? Would it really be so hard to put some flash on the FPGA itself? There are some FPGAs that have internal config memory (i.e non-volatile FPGAs). Volatile FPGAs tend to be cheaper, have higher densities, and usually more features. External config memories are pretty cheap, where as adding a memory into a chip adds a lot of cost to it and takes up a ton of space (if I can get my hands on an image of the internals of our ASIC I'll show you how much of it the 1M RAM takes up). Another reason is configuration. It's possible to have multiple images in the config memory and choose which one to load on power up. Among other things, this allows you to have a "golden image" to fall back on if an update is buggy or something.
|
|
#
?
Mar 16, 2009 14:15
|
|
|
ANIME AKBAR posted:okay, in all honesty you need to take a step back and do some more research on how amplifiers work. I can guarantee this circuit won't perform the way the simulator indicates, since the falstad simulator doesn't take a lot of things into account. It assumes that FETs will be operated under certain conditions, which you're not.  I was actually under the impression we were being eetrolled.
|
|
#
?
Mar 16, 2009 21:06
|
|
|
Delta-Wye posted:
The HC11 is the best microcontroller in existence. 
|
|
#
?
Mar 17, 2009 00:07
|
|
|
ANIME AKBAR posted:my university requires engineering majors take an intro to programming course, and when I took it all we learned was Java (pretty much useless for engineers). Apparently this semester they changed it so that it's broken into sections, some of which focus on labview, matlab, and VHDL. I can't believe I missed out on this stuff. My inability to use development environments like these is going to burn me later... How can you be such a genius at circuit design while knowing nothing about something as basic as programing FPGAs!?!??! Are you taking all of your engineering classes in reverse?
|
|
#
?
Mar 17, 2009 02:25
|
|
|
Corla Plankun posted:How can you be such a genius at circuit design while knowing nothing about something as basic as programing FPGAs!?!??! Are you taking all of your engineering classes in reverse? the FPGA class is a technical elective. There are lots of classes for digital logic, many of which involve VHDL and Verilog, but there's no practical component like in this course. And most people who are into pure circuit design probably don't need FPGAs if they're going to be designing power supplies or amplifiers or whatever. And I'd say the vast majority of my electronics know how has nothing to do with my formal education. No class here really teaches how to program microcontrollers, design pcbs, select proper components, or do anything really practical (besides a couple unique classes). I learned simply by pursuing them myself. If I know how to do something, it's because I needed to learn it in order to develop some project of my own, not because a grade depended on it. Hence my grades are utter poo poo, and a lot of employers probably won't even give me a chance.
|
|
#
?
Mar 17, 2009 02:54
|
|
|
Corla Plankun posted:How can you be such a genius at circuit design while knowing nothing about something as basic as programing FPGAs!?!??! Are you taking all of your engineering classes in reverse? What do FPGAs have to do with analog circuit design?
|
|
#
?
Mar 17, 2009 03:29
|
|
|
ANIME AKBAR posted:the FPGA class is a technical elective. There are lots of classes for digital logic, many of which involve VHDL and Verilog, but there's no practical component like in this course. And most people who are into pure circuit design probably don't need FPGAs if they're going to be designing power supplies or amplifiers or whatever. I assume you were being eetrolled At my school, all the EE/CmpE had to take 1 intro to digital design class which used FPGAs. And I'm sure that many of the analog guys didn't bother to think about it after that class was over. Similar to most mandatory classes that everyone has to take but aren't interested in. And I would be surprised that your school doesn't have any design classes. I know mine had several, ranging from analog stuff to embedded systems. Generally they are junior/senior level though.
|
|
#
?
Mar 17, 2009 03:40
|
|
|
ANIME AKBAR posted:the FPGA class is a technical elective. There are lots of classes for digital logic, many of which involve VHDL and Verilog, but there's no practical component like in this course. And most people who are into pure circuit design probably don't need FPGAs if they're going to be designing power supplies or amplifiers or whatever. If the economy pulls out of the crapper enough for us to lift the hiring freeze by the time you graduate (and you feel like moving) I'll refer you to HR where I work.
|
|
#
?
Mar 17, 2009 03:41
|
|
|
|
| # ? May 21, 2024 22:18 |
|
|
I'm in need of a logic analyzer that can do up to 500kHz (fortunately I only need 2 channels, I'm analyzing a proprietary serial protocol). Can anyone recommend a PIC or AVR based solution? I've seen a few plans on the Internet but I'm not sure what's good. It needs to be able to store a lot of samples (at least a few seconds worth) and I don't know which design to pick. Thanks!
|
|
#
?
Mar 17, 2009 03:41
|
|
