There has been an on/off again requirement for networking a controller used for data acquisition. One of my assignments is to do a preliminary study on the best way to do this (translated cheapest in dev. time and product cost). I need some help answering this because I don't have time or budget to try all the possible solutions.
The requirements are I2C, possibly SPI, RS-485 or CAN (not decided yet), flash file system, and TCP/IP stack with http, e-mail, ftp, telnet, and maybe snmp. I didn't specify the processor because I see there are some very powerful 8 bitters out there (MicroChip, Rabbit, Zilog) that may work, and of course the Moto 5282 and ARM chips. I understand that there are some highly integrated x86's as well. I suspect all may work, but I would like to hear comments from anyone that has worked with these processors/manufacturers.
The operating system can be very simple, a RTOS, or Linux. There are a lot of considerations here. The most important that I can think of are cost, royalties, and development time.
So, I would appreciate any comments from anyone that has walked this road before and has some good insight on a good direction. This will save me a lot of time and my company some money. Thanks for your help.
I guess there is absolutely no way to not do the work required. Go through all the possibilities and sort out the properties, advantages and disadvantages. For example, I2C and SPI are not to be put on cable, they are local to the board or board system.
These are immeasurably broad requirements (what exactly is this "device" supposed to do? It sounds like a Pocket Myriad; i.e. a theoretical device that changes into any shape based on the user's current or future needs; Ethernet gateway, cryptosystem, dragon-slaying sword, etc). Cost-savings can only be achieved through specialization, and specialization can only occur once the task requirements are understood. A $19 Discman can't play DVD-audio disks. And are you going to make exactly three of these boxes, or one for every man, woman and ferret in the northern hemisphere?
For a project with no specification and furthermore no projected sales volumes, which is basically what you've presented here, a reasonable heuristic is to consider that development hours will be the most expensive component and practically ignore the hardware costs. And, since you'll probably have to make a lot of changes in the application layer, you should pick an underlying OS and hardware architecture that has ready-to-use interfaces for everything you want, so at least you are spared the horror of developing H/W interfaces and device drivers.
Since you are talking about an ultimately general-purpose product, you should choose an ultimately general-purpose architecture - Linux on x86 is an excellent choice, and royalty-free. You didn't specify any kind of realtime or reliability requirements, so I'll assume you don't need to go with "carrier grade Linux" or any of the other buyware variants.
If I was given the assignment of "make a universal connector", I'd pick a PC to do it, just because the development time is simpler - well-understood tools, a plethora of free sample sourcecode available, [relative] ease of debugging, scalability, ...
Z-World / RabbitSemiconductor has Ethernet enabled SBCs and core modules that you can try in inexpensive development kits. There are no royalties for any SW. The full TCP/IP stack is included with the dev kits in source form, as are extensive samples, libraries for e-mail, telnet, SPI, RS-485, http, and a flash file system. A royalty-free version of the RTOS uC/OS-II is availables for $149, as is SNMP. CAN libraries are not included, but one of our SW people has written some CAN chip drivers already.
We are coming out very shortly with some propriety extensions to HTML and HTTP that work with compiler modifications to make the development of browser-interfaced embedded applications VERY easy. We'll be showing this new SW in the Z-World booth at the NMW show in Chicago next week. -Brian
If the requirement is "network enabling an existing product"...
Step 1 - do not roll your own. Unless you have an excess of time, are trying to be extremely tight on cost, and you're manufacturing a ton of these to recover development cost.
Step 2 - buy a module. These will run USD$35-$50 per unit and include everything you need - turnkey, customizable, with back-end interfaces.
I suppose these rule out using any form of Linux. If you are in that league, you might also take a look at Ubicom. Very cheap SOC (including RAM, Flash and versatile peripherals), good development tools (GNU based, but not free). 2K for smaller systems, 3K (no internal Flash) for higher performance and resource demand.
Rabbit seems to be good for small quantities, as they offer a complete multi-chip system. Easy to use but pricey.
The advantage of Microchip seems to be the free development tools (no idea about their quality) and the price for very small SOCs.
The Netburner MOD5282 has I2C, SPI, CAN, Uarts, Ethernet, HTTP,Telnet,FTP, E-Mail with SNMP and SSL as options. A Flash file system ported to the Netburner device is availible from HCC embeded. The development kits are releativly inexpensive and we have a 30 day return policy.
There is a lot to this request, so I'll help as best I can. First, the processor question. I have tried the Rabbit a year or so ago, but moved to the Zilog eZ80 because it has 2 UARTS, and that plays an important roll in this project. From what I have seen and can tell based on information on the web sites, the Motorola ColdFire is the king of the hill as it has everything but the kitchen sink. It is
32-bits, which may be a cost or packaging issue. NetBurner seems to have a good kit, but I have not tried it. I don't have much information on the ARM, AVR, TI, or Maxim processors, so perhaps someone else can comment on what they offer. I have also looked at the products that add networking via serial port (RS232 or SPI). I like the Digi and Lantronix units for their small size, but they are only RS232, which may be a problem for your application. The others require a small card, which will be a packaging problem for this project. As for Linux, I have not done an embedded project yet, but remember that this is a 32-bit OS, so the processor has to be 32-bit. Again, this may or may not be a packaging and cost problem. In addition, Linux out of the box is not really a real time OS. There is a patch for that issue, but I don't know how well it works. Perhaps someone that has used Linux with the patch can comment. It is also worth considering if you really need an OS. Most of my HC11 and 8052 projects use a timer and hard code to do this. Not the most flexible way, but gets the job done in simple applications. Something you didn't mention is developer support. This is a lot more important that one may first think. I'm sure that everyone reading this has been some task that will save the company if it can get done is less than 6 months. So the developer looks around (like you are doing now), then gets a development kit. The assumption being that the kit will get you up to speed in a few days and you'll be ready to decide if this is the processor/OS for you, if you need the professional version of the development tools, etc. Unfortunately, you may end up with over 1500 pages of documentation, no clue where to begin, and to get help you have to register, then post questions on a web form (with cookies enabled), and then check back periodically to see if you got an answer. This is how Zilog does it and it is a pain. Some kit producers have a better system, but they may cost more. You have to evaluate the trade-offs. Still, it is very important that you consider this because it will affect your development time. Based on simply the support issue, I may suggest the Rabbit over the Zilog processor for two reasons. One is that I see that Brian of Rabbit/Zworld provided a response. This means that they are interested in such matters, which is very positive. They also have a book written by Jan Axelson, which may save you a lot of time getting started. I have not looked at the book, so I can't say for sure, but it is definitely a step in the right direction. I hope I have helped, but if I have confused you,then send me a note and I'll try again.
Sincerely, Dennis Zimmerman
--
If sending a reply you will need to remove "7UP".
You will definitly want to check out our latest SBC with Ethernet, the picoFlash controller;
formatting link
This new board with Ethernet breaks the $100 price barrier - adding control and TCP/IP connectivity to your product cheaper and easier than ever before.
Slightly larger than a credit card, we've implemented a 40MHz x86 microprocessor, ½ megabyte each of RAM and Flash memory, two serial ports, battery-backed clock calendar and socket for DiskOnChip. Also included on the board are 16 digital I/O lines, a watchdog timer, RS-485 and TTL serial port I/O, and LCD and keypad drivers.
The picoFlash single-board controller comes with preloaded DOS operating system and all the utilities you need to start development. Included in our low-cost development kit (ONLY $129!) is a full copy of Borland C/C++ and the WatTCP TCP/IP stack. Also included is the source code for a number of TCP/IP clients and servers including HTTP, TFTP, FTP, Telnet and PPP.
For RTOS support, consider
formatting link
or uC/OS-II
formatting link
There is also an option for CAN Bus support using CAN232, here is a link;
This is what I especially like about Ubicom. They have a very active developers forum where users and Ubicom stuff discuss any planning issues and development problems. Moreover they have a system of "trouble tickets" that are supposed to be handled in a priority schedule. Of course you are only allowed to the system when you bought their development kit ($3000).
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.