Category: AVR

The ATtiny 25/45/85 datasheet has an intriguing section about the “dead time generator” that I found a little confusing. A little practical example helped me to understand it. The code and logic analyser trace (made using the same analyser and client mentioned in previous posts) appear below. This is just an example to understand how it works. Real applications seem to be principally brushless DC motor control (pdf).

The setup below uses both A and B compare registers with the same compare value and applies some dead time to the B output so that the effect can be easily seen. I was also tempted to play around with the C compare register which sets the value at which the counter resets. Read the code comments for more…

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41

//set the top count give whole number percentage duty cycles
const unsigned char top = 99;
//40% duty cycle if top=99
const unsigned char compare = 39;
//prescale CLK/8, 8Mz clock and div8 prescale -> 1MHz tick -> appropx 10kHz output with top=99
const unsigned char prescaleTimer = (1<<CS12);
//prescale CLK/4.
const unsigned char prescaleDead = (1<<DTPS11);// div 8 = (1<<DTPS11) | (1<<DTPS10)
// with CLK/4 prescale and 8MHz clock the dead time is 0.5uS per LSB.
// Dead time is delay to rising edge of signal
const unsigned char deadHigh = 0x0F; //8uS dead time for OCR1B. Max 0x0F
const unsigned char deadLow = 0x08; //4uS dead time for /OCR1B
 
int main(void)
{
	//set data direction for output compare A and B, incl complements
	DDRB = (1<<PB4) | (1<<PB3) | (1<<PB1) | (1<<PB0);
 
	//setup timer1 with PWM. Will be using both A and B compare outputs.
	// both compares will be the same but only B will have dead time applied
	OCR1A = compare;
	OCR1B = compare;
	TCCR1 = (1<<PWM1A) | (1<<COM1A0); //Compare A PWM mode with complement outputs
	GTCCR = (1<<PWM1B) | (1<<COM1B0); //Compare B PWM mode with complement outputs
 
	//PLLCSR is not set so the PLL will not be used (are using system clock directly - "synchonous mode")
	//OCR1C determines the "top" counter value if CTC1 in TCCR1 is set. Otherwise "top" is normal: 0xFF
	OCR1C = top;
	TCCR1 |= (1<<CTC1);
	TCCR1 |= prescaleTimer; 
 
	//setup dead time for compare B. Note the prescaler is independent of timer1 prescaler (both receive the same clk feed)
	DTPS1 = prescaleDead;
	//DT1A is unset - output A has no dead time
	DT1B = (deadHigh<<4) | deadLow;
 
    while(1)
    {
        //do nothing
    }
}

End Stuff

Source code is also available from github.

All code is copyrighted and licenced as follows:

***Made available using the The MIT License (MIT)***
Copyright (c) 2012, Adam Cooper

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

This article builds on the previous one.

This experiment was stimulated by wanting to have a 128kHz system clock but still be able to use the ADC. Section 17.5 of the datasheet clearly says:

… requires an input clock frequency of between 50kHz and 200kHz to get maximum resolution. If a lower resolution than 10 bits is needed, the input clock frequency to the ADC can be higher than 200 kHz to get a higher sample rate. It is not recommended to use a higher input clock frequency than 1 MHz.

Section 17.8 goes on to say:

The ADC is optimized for analog signals with an output impedance of approximately 10 kΩ or less. If such a source is used, the sampling time will be negligible. If a source with higher impedance is used, the sampling time will depend on how long time the source needs to charge the S/H capacitor, with can vary widely. The user is recommended to only use low impedant sources with slowly varying signals, since this minimizes the required charge transfer to the S/H capacitor.

There is some information on the web, particularly about the limitations of ATtiny/mega for high symbol rate signal processing but I wanted to try for myself and gather some data. The questions are: how does precision vary as both frequency and impedance vary, especially outside the specified range. Given the information in the datasheet, both the comments above and the general description of the sample and hold circuitry, the worst performance should occur for high impedance and high frequency. It turns out this is observed but the story is a little more interesting.

The Circuit and the Code

This uses the same approach as the previous post.

The circuit is minimal and constructed on breadboard. ICSP from an AVR Dragon was fed into a header and left connected. A 100n cap was bridged from pins 4 to 8 over the IC. Three different potentiometers were used: 250k, 47k, 5k. 5V was supplied from the Dragon.

The same code is used with both a 8MHz and a 128kHz main clock – the fuses are set to use the internal oscillators – and the ADC clock frequency changed using the ADC prescaler; after taking an ADC reading, the prescaler is moved to the next higher division factor. See the code below.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19

void Degrade_2(){
	unsigned char result[2];
	for(unsigned char prescale=1; prescale<=7; prescale++){
		//clear then re-assert the prescaler
		ADCSRA &= 0xF8;
		ADCSRA |= prescale;
		//start a conversion
		ADCSRA |= (1<<ADSC);
		//wait for end of conversion
		while (ADCSRA & (1<<ADSC));
		result[1] = ADCL;// datasheet says read low first
		result[0] = ADCH;
		sendBytes(result, 2);
	}
 
	//send a comma to separate readings
	unsigned char comma[]=",";
	sendBytes(comma,1);
}

Results 1: Low ADC Clock Frequency

This was a surprise. With a 128kHz main clock, the ADC gets clocked at from 64kHz down to 1kHz. No degradation of precision was observed, no matter which potentiometer was used and no matter what input voltage was selected.

Everything looked normal on the waveform, with the decreasing ADC clock rate clearly showing up as increasing conversion times.

So it looks like I can just use a slow clock and not worry about the ADC.

Results 2: High ADC Clock Frequency

For a 8MHz system clock the ADC clock will be 4MHz, 2MHz, 1MHz, 500kHz, 250kHz, 125kHz, 62.5kHz according to the prescale value.

A representative range of input voltages were used for each potentiometer using a multimeter which was disconnected before taking ADC readings. The ADC error is calculated assuming the 62.5kHz reading is correct. Three or more readings were taken for each ADC clock rate to confirm stability; no more than 1LSB variation was observed and the median was used. In all cases 10 bit conversion results are considered.

Although 4MHz looked OK sometimes, it is clearly very messed up! The effect of higher input impedance is clear but even so, we are getting 8 bit precision for most of the input voltage range at 500kHz. Remember the preferred input impedance is around 10k. Just outside the datasheet max freq, at 250kHz the error is down to the least significant bit. So it looks like this device at least is capable of adequate performance a little bit outside both impedance and frequency ranges.

The second plot shows more clearly the change of error as the potentiometer is swept across its range. The 3V readings seem to be particulary badly affected by the very high frequencies.

With a 47k potentiometer, which I consider to be compliant with the datasheet impedance requirement, the results look rather impressive. Even at 4MHz there is only a 3LSB error, i.e. we are getting 8 bits of precision (the top 8 bits of the 10 bit results) give or take some quantization error. Although… I am still rather suspicious of 4MHz and I did not explore a wider range of input voltages. At 2MHz, though, this device looks reliably better than 8 bits across most of the input voltage range.

The final plot shows the case for 3V, which looks like the worst-case, and neatly summarises how far you can push this ATtiny85. The 5k potentiometer is not really significantly better than the 47k except for 4MHz.

End Stuff

Source code is also available from github. A spreadsheet of results and the plots is also there.

All code is copyrighted and licenced as follows:

***Made available using the The MIT License (MIT)***
Copyright (c) 2012, Adam Cooper

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

As a precursor to investigating the precision of the AVR analogue to digital converter (on an ATtiny85 but assumed to be similar across many AVR devices) outside the recommended ranges of conversion frequency and input impedance, I set about to get to know the ADC better with a couple of “elementary” examples:

• a simple read in a while(1) loop
• a read triggered by Timer0

A Preliminary Diversion – observing the process with a logic analyser

There seemed like three main ways to see what the results were:

1. transmit the results using the serial interface provided on the ATtiny and capture with a logic analyser
2. as (1) but capturing the bytes on a PC (etc)
3. direct display using an LCD or LEDs

#1 has the benefit of allowing inspection of the timing as well as capture of results. I have an “Open Logic Sniffer“, which is a great bit of kit for getting to know your MCU and it is a bargain (although has a few minor oddities), which I use with the Logic Sniffer Java Client. The OLS client has some nice analyser features. This was my choice, not least because I had zero experience with the ATtiny Universal Serial Interface (USI).

#2 sounds OK but the USI isn’t quite as universal as it might be – no USART – and I couldn’t be bothered to set up an arduino to relay data, although that is on my “to do” list. Also, I did want to watch the timing.

#3 looked like too much effort on an 8 pin device, given the objective.

Given my zero experience with the USI, I opted for a 3-wire setup that gives a signal that can be understood as SPI; the ATtiny plays the role of a master and just blasts out bytes assuming there is nothing to receive. The OLS client can decode the signals and serve up the transmitted bytes.

 
First we must setup the USI and data direction. Note that “DO” is the data out line but that Atmel have given this the synonym “MISO”, which makes sense if the ATtiny is a slave or is being programmed with an ICSP. PB0 is used as a “slave select” signal, which makes for easier interpretation of the signal traces in OLS, both by humans and the SPI analyser.

1
2
3
4
5
6
7
8
9
10
11
12
13
14

//values to set USICR to strobe out
const unsigned char usi_low = (1<<USIWM0) | (1<<USITC);
const unsigned char usi_high = (1<<USIWM0) | (1<<USITC) | (1<<USICLK);
 
//setup pins for serial
//  PB2 (SCK/USCK/SCL/ADC1/T0/INT0/PCINT2)
// 	PB1 (MISO/DO/AIN1/OC0B/OC1A/PCINT1)
// 	PB0 (MOSI/DI/SDA/AIN0/OC0A/OC1A/AREF/PCINT0) - will not be used as DI
//	PB0 used as /CS
DDRB = (1<<DDB0) | (1<<DDB2) | (1<<DDB1);// /CS, USCK and DO as outputs
PORTB |= (1<<PB0);//slave not selected
 
//setup serial comms - 3 wire
USICR = (1<<USIWM0);

The following is called whenever one or more bytes should be sent out.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32

//sends the specified byte as serial (Three wire style).
void sendBytes(unsigned char bytes[], unsigned char size){
	//slave select
	PORTB &= ~(1<<PB0);
 
	//loop over bytes
	for(unsigned char b = 0; b<size;b++){
		//load the byte to the output register
		USIDR = bytes[b];
		//strobe USICLK 8 cycles (also toggles clock output thanks to USITC)
		//an unrolled loop gives 50% duty and saves 3 clock cycles per bit sent
		USICR = usi_low;
		USICR = usi_high;
		USICR = usi_low;
		USICR = usi_high;
		USICR = usi_low;
		USICR = usi_high;
		USICR = usi_low;
		USICR = usi_high;
		USICR = usi_low;
		USICR = usi_high;
		USICR = usi_low;
		USICR = usi_high;
		USICR = usi_low;
		USICR = usi_high;
		USICR = usi_low;
		USICR = usi_high;		
	}//bytes loop	
 
	//slave de-select
	PORTB |= (1<<PB0);		
}

Testing this with a “Hello World!” message

		unsigned char send[] = "Hello World!";
		sendBytes(send, sizeof(send)-1);	//-1 drops the null

and running through the OLS Client SPI Analyser tool gives:

This is “mode 0” SPI style; see that the data is shifted out on a falling SCK and sampled on a rising edge. The alternating high and low assignments to USICR give a USI clock period 1/2 of the main clock since each assignment is a single cycle operation. Both can be seen in the following trace, which shows b01100100 being shifted out.

General Structure of the Code and Other Notes

Inside main() I first do some setup then I have a while(1) loop inside which is a function call. Each example/experiment exists as a separate function. Although this leads to the overhead of a few cycles to call the function, this is convenient in the present cases.

I also blew the fuse “CKOUT = [X]” so that the system clock is accessible to the logic analyser.

Simple Example

This assumes the fuse setting “SUT_CKSEL = INTRCOSC_8MHZ_6CK_14CK_0MS”, i.e. an 8MHz internal clock. Make sure that the logic analyser sampling rate is 20MHz or higher otherwise the clock signal will be mis-captured.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16

//assumes 8MHz clock, hence prescaling 
void SimpleRead(){
	//set prescaler to div64 --> 125kHz ADC clock
	ADCSRA |= (1<<ADPS2) | (1<<ADPS1);
	//start a conversion
	ADCSRA |= (1<<ADSC);
	//wait for end of conversion
	while (ADCSRA & (1<<ADSC))
	{
		//do nothing
	}
	unsigned char result[2];
	result[1] = ADCL;// datasheet says read low first
	result[0] = ADCH;
	sendBytes(result, 2);
}

Since this waits until the ADC conversion has finished, the interval between readings is effectively controlled by the time it takes for conversion. This is, in turn, determined by the ADC clock. The datasheet indicates the ADC clock should operate at between 50kHz and 200kHz. The datasheet also says a conversion takes 13 ADC clock cycles. Given the prescaling (code above), the conversion should take 13*64 = 832 cycles. The logic analyser (no screenshot shown) shows 901 cycles between /CS rising (happens at the end of sendBytes just before returning) and /CS falling again (happens just after entering sendBytes on the subsequent reading). Hence there are 69 main clock cyles spent doing other things: returning from sendBytes, returning from SimpleRead, process the while(1), call SimpleRead, setting ADCSRA, processing the while loop in SimpleRead, retrieving the low and high bytes, calling sendBytes. There is clearly some waste here that should be avoided in many real applications.

Timer-triggered Example

This is a rather more interesting example in which Timer0 is configured to trigger ADC conversion when the timer value = the “compare A” value. During both the timer delay and the ADC conversion, software is free to do other things since timer and ADC are hardware-controlled. At the end of the conversion an interrupt is triggered to do something with the result. In this case, just to send it out to serial.

Note that this example uses the 128kHz “watchdog” clock (fuse: “SUT_CKSEL = WDOSC_128KHZ_6CK_14CK_0MS”) to give a long enough timer 0 interval for me to detect the delay. I also used a 2s delay to change the ADC input during execution and make sure it is being properly captured. This means I had to reduce the ISP clock for in-system programming; I used 16kHz, which is reliable if slow.

The code is as follows, noting that Init_TimerTriggered() need only be called once. Note also the last line in the interrupt service routine… that one took me a while to work out!

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30

//setup for TimerTriggered
void Init_TimerTriggered(){
	//1. enable ADC completion interrupt
	sei();//global interrupts
	ADCSRA |= (1<<ADIE);//ADC interrupt
 
	//2. set the ADC to be timer triggered
	ADCSRB |= (1<<ADTS1) | (1<<ADTS0); //this defines the trigger source
	ADCSRA |= (1<<ADATE);//this is needed to enable auto-triggering
 
	//3. setup timer
	TCNT0 = 0x00;//counter to 0
	TCCR0A =  (1<<WGM01);//use "clear timer on compare match" mode
	OCR0A = 0x80;//output compare to 128 gives about 1s with 128kHz sys clock and prescaler (below)
	TCCR0B = (1<<CS02) | (1<<CS00);//prescaler to 1024, which enables the counter
}
 
//ADC completion interrupt service.
//Sends the data from the ADC register
ISR(ADC_vect)
{
	//read and send ADC
	unsigned char result[2];
	result[1] = ADCL;// datasheet says read low first
	result[0] = ADCH;
	sendBytes(result, 2);
 
	//clear timer compare flag otherwise the ADC will not be re-triggered!
	TIFR |= (1<<OCF0A);
}

Given the slower clock, the logic analyser sample rate and total capture package can be reduced too. I used a capture trigger (type=complex mode=serial in OLS… read the OLS tutorial!) to watch for /CS falling so 1kB of capture data at 500kHz sampling is ample to capture one timer-driven event.

The captured trace (following ADC completion), just for completeness is:

End Stuff

Source code is also available from github.

All code is copyrighted and licenced as follows:

***Made available using the The MIT License (MIT)***
Copyright (c) 2012, Adam Cooper

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.