Years ago when corporations came into being some management pattern was needed.
As there was no business model to copy something had to be adapted. Which organisation have to manage many people and work on complex projects? Army.

In Army there is whole command chain with president or a minister on the top, some generals and captains to work on maps and papers, sergeants and corporal to work with soldiers in the field. And of course the team to kill bad people whatever bad means.
In corporation minister is the CEO/chairman, generals and captains are the upper management to negotiate contracts, create and cancel projects. Sergeants are lower management – team leaders and all other X leaders – they work with the team close to the project. And of course the team to kill… I mean to do the job. Whatever job means.

After years of working with army style management corporations try to be agile. Less management, no X managers, self organising, cross functional teams.
Cross functional teams are quite popular also in army. Team with about seven people with different skills and armament could do a big mess… Isn’t it similar in scrum? No separate testers and developers team in scrum, no separate launcher teams and M16 teams in army :-)).

How might army look like if they use scrum?

Beginning
Sergeant: Team 167, you have to attack the tower…
Team 167: Serg, you have no longer the authority over the team, sir. We will choose our targets after meeting with Captain… I mean Product Owner.

Sprint planning
Captain: Teams, last sprint no team decide to attack the tower. This is the item of high priority and you must do this during this sprint.
Teams: No team volunteered for this item because it’s too big and not clarified enough. Teams would like to know if there are snipers in the tower and how many nades enemy has. Last time we encountered many impediments because items were not specified enough – land mines and one dangerous ambush. Captain, you have to work more on items specification.

Sprint – Daily scrum 1
Sergeant: Team 167, Captain wants us to rescue the hostage. We can do this on our way to the tower.
Team 167: Captain is not allowed to add new items during the sprint. You should know this as a scrum master… By the way, we don’t want you to be our scrum master any longer. You don’t obey to the scrum rules.

Sprint – Daily scrum 2 (sergeant got his last chance :-))
Sergeant: Team 167, I want to remind you KABOOM that you should meet daily. It’s very important to synchronise work between team members and… DADADADA BOOM plan next activities…

Sprint review
Team 167: Captain, we manage to kill enemy and occupy the tower.
Captain: Fine, so item #1352 is finished. Good job soldiers!
Sergeant: Sir… As a scrum master I have to remind you our definition of done. Tower is not jet fortified, tower has no communication with our base and the road to the tower is still mined.
Captain: Ok, so I can’t mark this item as done. Teams please bear in mind that item #1352 has high priority and must be finished until hmmm… release.

Can you imagine? :-)

Looking for a job as C++ developer? Waiting for another job interview? Tired of “what does encapsulation mean?” questions and being judged after half hour interview? Then try to judge your potential employer.

Of course first part of job interview is always pretty much the same and there are many howtos about handshaking, eye contact and such stuff. Then you have more or less known questions and here is first interesting part.

Beginning is usually simple – where did you work before plus some basic C++ questions. This definitely shouldn’t be the last part. If programming language basis is enough to get the job, and it wasn’t interview for greenhorn, newbie or trainee then you should run away. Really.
“Our system architect knows even how to overload operators in C++”. Dude, run faster!
Even if there was “young but very experienced team” statement in job announcement. Such experience might be painful to share.

Definitely there should be more questions about design patterns, code architecture, uml and tools. If you get this job you’ll probably work with interviewers or people with similar knowledge. If they are better then you – you’ll be able to learn something new.

If everything went well interviewer lets you to ask questions. Think before the interview what could you ask about. All questions about holidays planning, salary, bonuses, social things are important. However don’t forget about technical things.

My favourite is “Do you use stl?”. If the answer is “no” then listen to explanation very carefully. If they work on something with operating system then there is probably no explanation. If you’re told that stl is not efficient, that they developed better and faster library on their own then thank them for their time and flee. Yes, there still are people who reinvent the wheel and poorly rewrite the stl library. We don’t call them morons only because we are well mannered.

More good questions – “Do you use boost?”, “Do you support open source?”, “Do you use any open source libraries in your projects?”. If they don’t want to use free third parties and want to rewrite whole universe their own “better” way then you know what to do.

Of course until you have family to support :-)).

I’ve changed my job about two months ago and now I work in Nokia Siemens Networks.

I can’t say bad word on my previous employer – Sonel. Interesting projects and great team, but I just had to try something new.

So now, after one and half year of low-level-but-still-object-oriented programming I’m working on embedded rtos.

By the way – I would like to thank you for your emails – some people asked me for ATMega + CC1000 code. Well I’m very sorry, but it’s not possible to put this project here. First thing is that I should rewrite, test and document part of code and I just don’t have enough time to do this. Second is that I don’t have hardware from this project – it’s property of my University.

It would be really nice to run such project on more powerful microcontroller, but I’m not able do develop boards and port/rewrite code. Just because lack of time.

Hardware Abstraction Layer is a part of software that (almost) allows you to forget about hardware you’re working with. In perfect case you can use the same software on different hardware just by changing the hardware layer.

Clean and well designed HAL hides hardware differences and separates software layer responsible for hardware managing from main application functionality.

There are some issues to consider before designing and coding HAL…

Memory and time…

Simple example below. We need some text debug output working on rs232 in our application.

C_Debug class uses pointer to C_Uart object. C_Uart is an abstract class, part of Hardware Abstraction Layer. Because Send and Receive are pure virtual methods C_Debug class doesn’t know what is the exact uart peripheral. If you implement your own C_Uart<microcontroller_name> class with Send method, you will not have to rewrite software which uses C_Debug.

Of course it’s not so perfect. Virtual methods aren’t so cheap – when you call virtual method there has to be information about exact implementation that should be used. Calling Send method by C_Debug (on pointer pUart) is in fact executing some extra code before executing proper method. Choosing right method is done by checking the virtual table, which should be placed in flash memory.

Of course above uart example is very simple, but imagine bigger HAL with many peripherals – it may cost much memory (vitables) and it may make difficult executing fast functions with exact timing. There are (of course) methods to avoid delays but most of them are more like workarounds then clean solutions.

In Uart example (from this blog) you may see two namespaces – Devs and Periphs. Lets pretend ( :) ) that Devices uses objects from Periphs namespace. The easiest and cheapest way to get portable, HAL alike solution is to cut periphs.h with defines.

#include "uprocdefs.h"
#ifdef UPROC_ATMEGA88
    #include "HAL/ATMega88/uart.h"
#else
    #ifdef UPROC_HC11
        #include "HAL/HC11/uart.h"
    #else
        #error No microcontroller definition!
    #endif
#endif
namespace Periphs
{
    /// UART peripheral.
    extern C_UART UART0;
};

Of course there must be definition of C_UART class in both files and you’ll have to compile proper .cpp file in your project. It could be done in Makefile.
From now on you may use UART0 object from namespace Periphs. There is no virtual methods, C_UART classes from HC11 and ATMega88 directories don’t even need same base class. If C_Debug class uses Send method and this method (same name, parameters and return value type) is implemented in both files this example will work.

Small project in KiCad.
ATMega88 with power supply, isp and rs232 connectors on first schematic.

Here in png file.

Module with max232.

Here in png file.

It’s good idea to make separate, small module for rs232 communication – it may be useful in other projects. You should think about rs232 connector pinout (between board and module). In this case it is possible to plug P2 (RS232) inversely – this may burn up your rs232 module and even com port or supplier.

I haven’t got max232 for 3.3 V power supply so I usually put 5 V regulator on rs232 module. As far as I remember ATMega88 tolerates 5 V on its inputs even when it’s powered with 3.3 V.
To launch this board you’ll need also stk200 programmer, it’s very simple and easy to google.
On this basic board you may run uart example. Of course you need to configure clock speed (fuse bits + uart registers)…

ATMega88 basic board – KiCad project.
AVR ATMega88 – KiCad lib.

I’ve been moving to our new house for two months and I still need one month at least to finish renovation. Because we’re doing most things without any help it’s really time-consuming. Moreover I’ve got peak month at work, we have some new hardware. I know that every man is a boy and I have to say that getting some new devices, is for me like getting favorite candy. Especially when some (even small) part was designed by me. Working with new hardware, getting it to work, putting small functionality together into bigger pieces just make me happy.

Anyway – I had one-week holidays :-) in Poland and Czech Republic. What can I say… They have much better roads. Weather was quite good – nights were sometimes rainy days usually cloudless (in fact I don’t afraid of rain, right..?).

I’m not going to complain or something, but selling alcohol in small towns and villages in my beautiful country should be forbidden. It’s sad that you meet the same drunk morons in front of only shop in the village every day. Fortunately you may still meet some good people and find charming places, usually with no people at all.

NXP published new erratasheet for lpc23xx family (see previous post).
It’s available here.
To summarize – sometimes turning MAM fully enable may cause instability. Workaround – don’t (fully) enable MAM.
I’ve expected much more details, so I’m really dissapointed. Maybe they aren’t sure yet when and why exactly MAM fails…
Anyway I haven’t noticed new errata for lpc21xx family. Maybe they mismatched family name with lpc24xx, which of course is affected by this bug. “Of course” because lpc24xx is based on lpc23xx.
There is one more bug about code execution speed. If code is executed from flash memory frequency is limited to 60MHz. I’m not sure what it means. Is 60MHz maximum main clock frequency or maybe for clock faster then 60MHz code will be executed with same speed? In first case – peripherals and code work with max 60MHz. In second – code is executed as with 60MHz clock, but peripherals (like timers…) may work up to 72MHz.
Far for now – with 72MHz clock and MAM disabled I consider lpc23xx working stable.

Is this good news for NXP lpc23/24xx users..?
lpc2000 mailing list – lpc speed problem
In short – there is some issue with MAM (Memory Acceleration Module) in LPC23/24xx. If microcontroller works with high frequency clock (max is 72MHz), the MAM is enabled and there is more then 32KB of code, it may work unstable. Of course when MAM is disabled everything works just perfect. Funny thing is that same code may work perfect on another lpc microcontroller or even on the same after reflash. Same thing with frequency – you may find some information about solving MAM problem with changing frequency below 60MHz. There are also some posts about MAM problems when code is bigger then 32KB, but I think problem is related with code location, not size.
Issue may be also connected with PLL problems in lpc23xx family.
Anyway – I’m quite surprised that NXP (if guy is really from NXP…) noticed this bug so late, but better late then never. It’s hard to belive that lpc210x family has same bug (has NXP been selling it for more then a year..?).

Workaround proposition sounds like some voodoo magic – insert some NOPs and limit frequency. I would add “pray” :-).

After small reprimand I have to say that their microcontrollers even without MAM are quite fast – 72MHz is very nice frequency in ARM7. The thing is that with acceleration enabled it works even 4-5 times faster.

It’s very good that NXP finally admitted that problem exist. I hope new errata, with more precisely explanation of this bug, will be available soon.

Here is some interesting posts about MAM problems in lpc23xx family:
Investigation of LPC23xx MAM Issues – lpc2000 mailing list discussion.

Low frequency main clock…

If you use simple ISP programmer (like stk200) and your AVR microcontroller stops responding after changing clock frequency to very low – don’t panic.

I’ve got similar problem with avrdude – it just can’t detect atmega88 after I change clock speed to about 100kHz. Because I’m using the power save mode (those with timer2 clock enabled) changing the main clock prescaler gives me about 100-200 uA less current consumption. What I tried to do was to change system prescaler before the sleep mode (I’ve changed timer2 prescaler so timer worked with same speed in normal and sleep mode).

Avrdude can’t detect device because it’s to inpatient :-). All you have to do is talk to your AVR a bit slower. I’m afraid it can’t be done with avrdude (at least you probably may put some delays into avrdude code to make it work slower, avrdude is an open source utility, so use the source luke :-) ). Theoretically AtmelISP had some options to change frequency of programming avr with stk200, but I could’nt find any working with atmega88 and stk200 aplication on Atmel site.

Use uisp!

Uisp is an isp programmer, like avrdude. I’m not sure if uisp supports newer avr microcontrollers (I’m not so sure if “new” is the right word here) like atmega88, but I was able to erase flash memory with atmega8 parameter. After that avrdude starts detecting my atmega.

I’ve taken on to much things. Daily full-time job (which I shouldn’t write about) with some interesting new ARM microcontroller, motorcycle season approaching, some private things (which i definitely shouldn’t write about ;-)) and very short (three days) vacations…

Anyway I’m still alive. I’ve bought acceleration sensor MXA2500 and I’ve made it working on some prototype board with Atmega88. I’ve had to order new soldering tool, the old one is damaged (it was crap anyway). I’m trying to work with free Eagle software under Linux (to be frank – Eagle + qt looks ugly) so I hope you’ll see some simple diagrams here soon.

I’m starting work on my motorcycle alarm as soon as I get new tools. First I’ll put here some quickstart guide with simple prototype board diagram (ATMega88, some leds and MAX232 for UART), few words about AVR programming under Linux and Windows and some cruel facts about fuse bits. On this board you’ll be able to run example from previous post (Mega88, UART, C++).

In next step I’ll put here diagram (based on previous one) with MXA2500 and software which detects movement (pull or tilt).

Then I’ll have to think about connecting some noisy horn, make some “user interface” for turning the alarm on and off, power saving and some box (waterproof of course ;-)) to put everything into it.

After that point it could be only better, so maybe remote control, panic, notifying and so on…

So… stay tuned :-)