Filtering Data

Before displaying the data, some sort of filtering is usually required that removes fluctuations in the weight display caused by noise and vibration, but filtering comes at the cost of response time. The raw samples from the ADC take 100ms to acquire, so any form of low-pass filtering will slow down the response time into the region by fractions of a second. In the case of a workshop balance, a couple of seconds for the reading to settle is not a problem, but in other cases (e.g., a batch-weighing application), every second adds to production time and therefore costs money. Filtering and ultimate accuracy will always be a trade-off. In this application, 10 samples are simply averaged before the result is passed for display (Listing 3).

01 #define SAMPLES 10
02
03 int32_t samples[SAMPLES];
04 int32_t average = 0;
05 uint8_t sample = 0;
06
07 // main acquisition loop
08 while (true)
09 {
10   samples[sample++] = ADS1232_Read();
11   if (sample == SAMPLES)
12   {
13     sample = 0;
14   }
15
16   average = 0;
17   for (uint8_t i = 0; i < SAMPLES; i++)
18   {
19     average += samples[i];
20   }
21
22   average /= SAMPLES;
23
24   // pass the averaged data to the display here
25 }

Practical Precision

In terms of practical precision, you must keep a few things in mind. Of the theoretical 24 bits of precision available from the ADC, the datasheet itself admits that only 19 bits are useful at the highest gain setting. The load cell I have chosen has a maximum output of 4mV (1mV/V, with an excitation of 4V), so it uses only one fifth of the available range (20mV) of the ADC. That's about 100,000 counts. Practically, I have found I can achieve stable readings to 10,000 counts. For a 100g scale, that means one count is equivalent to 10mg. The current software limits that further, giving a resolution of 100mg (i.e., 1,000 counts), which seems more than adequate for the requirements of a general-purpose laboratory scale. There is certainly scope for better filtering and other techniques to improve the precision, but at some point the limitations of the (cheap) load cell will become the limiting factor.

Display Driver

The display has a three-wire serial interface. The chip select (CS) line must be asserted before data can be written, and once that is done, data is presented on the data (DAT) line, being clocked into the chip by asserting the write (WR) line and then negating it (Listing  4). The interface is too slow to keep up with the microcontroller at full speed, so busy-wait loops are necessary to slow the WR line pulses down to just over 3µs.

01 /** */
02 static void LCD_WriteBit(bool bit)
03 {
04   // set the data line
05   HAL_GPIO_WritePin(LCD_DAT_GPIO_Port, LCD_DAT_Pin, bit);
06
07   // assert the WR line
08   HAL_GPIO_WritePin(LCD_NWR_GPIO_Port, LCD_NWR_Pin, 0);
09
10   // busy wait
11   for (uint8_t i = 0; i < 50; i++)
12     ;
13
14   // negate the WR line
15   HAL_GPIO_WritePin(LCD_NWR_GPIO_Port, LCD_NWR_Pin, 1);
16
17   // busy wait
18   for (uint8_t i = 0; i < 50; i++)
19     ;
20 }

The display chip has a small amount of internal memory, and the individual bits in this memory correspond to the different segments in the display characters, as well as decimal points and the battery state indicator. Each transaction with the chip is called a command, and the first three bits of that command determine the type of command. The rest of the transaction comprises an address in the internal memory and the data to be written.

The process of writing a complete digit involves conversion from the required ASCII digit to the memory bit pattern to illuminate the display correctly. This code is not complex, but quite long, so I won't include it here. I refer the interested reader to the source code on my GitHub page [11].