My experience with programming embedded systems can be divided into two parts: sensor devices and our testbeds.

• Sensor Networking Applications

My main platform for research projects and papers is coding in nesC language for MICA2 devices. nesC is an extended C based language which supports event triggers.

The resulting image from the above code is one single binary file that includes all the drivers to interface underlying hardware. I sometimes have to change the way those drivers work, for example when I am altering the MAC protocol. Per usual definitions, this is a pure “firmware” for MICA2 devices.

I also have some experience with coding firmware for Ember 2420 based sensor devices in C++. However, this is not my main platform of work and I do not consider myself an expert on those.

• Testbed Design and Implementation

For ASSERT testbed our hardware was a custom platform consisting of a Lattice FPGA and an i.MX27 Freescale processor and some peripherals. A custom RF board would control units under test. We wrote distributed software in C, running on the processor that controlled the RF board’s functionality through the FPGA. Control PC part was written in C++ and Java. Nodes boot using TFTP and RedBoot.

We had to develop and maintain the Linux driver. For the prototype our code worked on Atmel’s NGW100 mkII, an AVR32 microprocessor based development kit. I was one of the three Ph.D. students who designed ASSERT’s architecture and oversaw software development.

Successor to ASSERT, we are designing WiNeTestER. This generation is capable of larger frequency range, phase-shifting and other features. Hardware for this phase is based on and FPGA board, WARP. Since we do not have a separate processor, most of the computation will happen on the PowerPC “on” the Virtex-4 FPGA.

In this phase I had to learn some primitive FPGA design to create my own reference design (due to some licensing issues and incompatibilities). Nodes will boot using TFTP and U-boot. I am one of the two Ph.D. students who work on the software aspect, and cooperate with our hardware team on the design requirements.

For both testbeds, we cross-compiled the Linux kernel for the processors (AVR32, i.MX27 and PowerPC) and created a driver for our custom hardware. Embedded part of the code is written in C and cross compiled as well. As a result, this part is not pure “firmware” and is more like porting our own code to different architectures. The control logic, running on a control PC, is written in C++ and Java.

ASSERT is fully functional and used on a daily basis. WiNeTestEr, funded by NSF, is currently being developed.

One would expect that over the years this was solved. This week I was adding new virtual servers to our Vsphere “cloud” and guess what! Same problem still exists (or at least I feel it’s the same).

It *is* a bug, not a feature! Please fix this!

Instead of changing the variable and recompiling, you can do the following. Declare your desired variable(s) as

extern const uint8_t myVar;

After compiling the main image, take a look at your symbol table by typing

 tos-set-symbols --read /path/to/main.exe

This will show a table like following:

AMQueueImplP__0__current             :  01  [.]  (1)
AMQueueImplP__1__current             :  01  [.]  (1)
ActiveMessageAddressC__addr          :  01 00  [..]  (1)
ArbiterP__0__resId                   :  01  [.]  (1)
CC1000RssiP__currentOp               :  04  [.]  (4)
CC1000SendReceiveP__BIT_CORRECTION   :  1b 1c 1e 20 22 24 26 28  [... "$&(]
CC1000SendReceiveP__rxBufPtr         :  5c 02  [\.]  (604)
CC1000SquelchP__clearThreshold       :  20 01  [ .]  (288)
MyModule__myVar              :  00 00  [..]  (500)
SerialDispatcherP__0__messagePtrs    :  fb 02 2a 03  [..*.]  (53084923)
SerialDispatcherP__0__receiveBuffer  :  fb 02  [..]  (763)
SerialDispatcherP__0__recvType       :  ff  [.]  (255)
TOS_NODE_ID                          :  01 00  [..]  (1)
atm128_stdout                        :  00 00 00 02 00 00 00 00 fb
00 00 00 00  [..............]

tos-set-symbol /path/to/main.exe MyModule__myVar=1234

This allows you to directly set the value of your desired variable. Just be careful, as any programmer should be, about variable types.
tos-set-symbols is part of the avr-gcc environment, and uses avr-objdump.

Initially we used this method to generate new images with different node IDs (TOS_NODE_ID variable) from the same base image. Now our testbed’s web based interface allows the user to draw a topology visually and assign such variable values to each of the nodes. It can also automatically map your desired topology to available nodes in the testbed.

Nice view Thank goodness I don’t have exams anymore.

In order to promote their main socail network, Google Plus, they eliminiated Reader’s features with no provision of seemless migration. I can see this as an engineer’s assumption that new “toys” automatically attract new people. Marketing people might disagree, pointing to brand loyalty.

The engineering versus marketing approach Google took towards Reader reminds me of their Android operating system. In October Michael DeGusta of The Understatement summarized his observations about Android updates for devices shipped between 2008 and 2010 as a chart. His visualization shows a clear message: Google does not intend to support their older devices.

My own experience supports this horrifying observation. While I am not sure about internals of Android’s framework and drivers, the ending result for users like me has been shaky reliance on courtesy of application developers to include my device in their small list of supported devices.

My experience with iPhone and Windows Mobile/Phone have been clearly different. The fact that I still receive updates for my 4 years old device says a lot. But the problem with iOS is it only has one active line of devices. You can get a newer version of iPhone but that is your only choice. However, Windows Phone has been more welcoming. My observation was WP7 suffered a poor release process, including poor support of developers in the early days and taking months before releasing copy & paste capability in an update. I hope they don’t make the same mistake again, and I can safely return to Windows Mobile family!

Update: I ran into another article discussing how internal instant messaging products of Google are not “synchronized” with each other. In my opinion it shows lack of a unified standard layer between users and Google products. As a friend pointed out, Google has been more concerned about shipping new products faster rather than keeping them synchronized.

But after a one semester trial, I am completely disappointed. Apparently Outlook is not intended for use with an IMAP server. Each time it would take almost minutes to load messages. For Thunderbird, it was merely seconds on the same server.

Apparently Microsoft developers are aware of this problem, but it is not fixed yet. There are hacks and suggestion to improve it, such as this one. Unfortunately these “solutions” don’t fit my needs and I am back to using good old Mozilla Thunderbird.

This announcement comes few days after AT&T made public they are considering cutting back amount of data users can transfer on their iPhones. AT&T has been criticized for network problems due to higher data usage, and now seems to be projecting the problem to their customers. You might be wondering how much data on average an iPhone user transfers per month? This article puts that number close to 100MB in 2008, although the source did not mention if that amount is per month or per day. Even if we assume it is per day, a very unlikely assumption, the total would be close to 3GB per month. Again, very unlikely, but let us assume that is the case.

The question that popped up in my mind after reading about the 4G mobile finally coming to the market, was how will the providers handle this new rates? If AT&T cannot handle current small number of iPhone owners, estimated to be around 6.4 million by Nielsen, how will they be able to persuade users to upgrade to 4G cell phones once they are available? The technology of the cell phone devices and the BTSs will make transfers quicker, but still the back bone that is required to carry this amount of data needs to be upgraded. And if they are more interested in “managing” how their customers use their cell phones rather than making that upgrade right now, there is no guarantee they will be willing to do perform those upgrades once customers start using the faster 4G devices.

I hope the major cell phone providers understand that when a customer pays higher price in terms of mandatory data plan and a longer contract, they expect to use the device they received as an incentive. And I hope while researchers and developers in hardware manufacturing companies are working on the technical aspects of the 4G devices, the business people in carrier companies are also working on understanding customer data usage trends and plan to upgrade their networks accordingly.

UPDATE, three months later: March 12 2010, AT&T CEO unsure if LTE will trigger pricing changes but “reiterated his position that such a move would happen.”

calling AMSend.setSource( *amsg,  addr) may change the source address locally, but when you are sending the packet it will be reset to ActiveMessageAddressC$addr.

After almost two days of wondering why I couldn’t forge messages on a base station, a friend more familiar with TinyOS gave me this warning. I couldn’t find it on the web, so .. here it is!

One other note: if you are using a bash script to do the TOS_NODE_ID assignment for you, you should call tos-set-symbol with ActiveMessageAddressC__addr. Using $ would -obviously- cause bash to consider it a variable and replace it with empty string.

In the recent years more and more wireless providers have started to provide femtocells to their customers. The customer can deploy the femtocell in any location that the coverage of normal towers are weak or absent.  Consider a product like Samsung’s Ubicell. The user can plug this femtocell in their home internet switch, and use their regular cell phone to connect to the final service provider, e.g. Sprint.

E-UTRAN architecture from 3GPP document 36.300-880

As of December 2008, 3GPP documentation for Long Term Evolution networks has started to mention Home eNB as part of the Evolved UTRAN. This move can be a significant advantage for the final users, only if service providers fully recognize the extent of service this final addition can offer. This recognition might require a careful study of 802.11a/b/g Access Points, considering the similarities of the two products. Both these products are the front end of a network to final customers. While these final points are accessible to the customers, the actual network is not controlled by them.

Most APs initially were merely wireless transmitters, but they soon started evolving into more sophisticated equipments. The need to provide more functionality for the end user to adjust settings, e.g. channel or user access rules, resulted in simple interfaces to configure the AP. Currently wireless APs are commonly acting as firewalls, switches, DSL modems.  But the true revolution for the APs happened when vendors such as Linksys started to provide open source firmware for their products. Because of the delay in this revolution, many golden opportunities are gone and we do not see as many diverse applications.

Given how larger organization such as universities might consider deploying a number of Home eNBs throughout their campus, there is a high level of interaction that would be only limited to these eNBs. As an example, in a university campus many calls and data interactions would be between students on the same area. Or consider an event like SXSW. While this year AT&T added enough resources to guarantee good coverage, smaller scale events do not motivate service provider to make such adjustments. Instead the organizing institute is well aware of the scale and the available resources and would be able to allocate more eNBs to the event area. One other aspect is access control: if I am paying for a Home eNB, I may not want my neighbor to user it for free unless it is an emergency call.

By providing good configuration options to the deploying organization, all sides of such scenario would benefit: the organization would enjoy faster service on their campus, the service provider would not have to worry about resource management and the manufacturer would ensure better sale of its products.

There are many challenges in providing such control to the final user. However, I believe the designers of 3GPP LTE have started taking initial steps and now the manufacturers need to move forward.

Related:

• Sprint Airave
• T-Mobile @Home
• Verizon Wireless Network Extender
• AT&T: 3G Microcell?
  
• Femtocells for Everyone