How to calculate baud rate register values for a UART
Divide the clock by the oversampling and the baud rate. An AVR uses UBRR = fOSC ÷ (16 × baud) − 1, so 9600 baud at 16 MHz gives UBRR = 103 and 9,615 baud, an error of +0.16 %. An STM32 uses BRR = fCK ÷ baud, so 115200 baud at 72 MHz gives BRR = 0x0271. Keep the total error across both ends of the link under about 2 %.
Baud rate formula for AVR, STM32 and generic UARTs
A UART divides its clock to get a sample clock a set multiple of the baud rate. Only whole divider steps exist, so most combinations of clock and baud rate land slightly off.
AVR: UBRR = fOSC ÷ (16 × baud) − 1 baud = fOSC ÷ (16 × (UBRR + 1))
- fOSC
- system clock in Hz
- UBRR
- baud rate register, 0 to 4095
With the double speed bit U2X = 1, replace 16 with 8.
STM32: baud = fCK ÷ (8 × (2 − OVER8) × USARTDIV)
- fCK
- clock feeding the USART, in Hz
- OVER8
- 0 for 16× oversampling, 1 for 8×
- USARTDIV
- divider as a fixed-point number in BRR
With OVER8 = 0, BRR holds 16 × USARTDIV, so BRR = fCK ÷ baud. With OVER8 = 1, the upper 12 bits hold the whole part of USARTDIV, bits 2:0 hold the fraction in eighths, and bit 3 must be kept clear. USARTDIV has to be at least 1.
Generic: baud = fCLK ÷ (S × D)
- S
- samples per bit, usually 16
- D
- divisor, with n fractional bits in steps of 1 / 2ⁿ
The error is the actual rate against the one you asked for, as the Microchip data sheet defines it:
Error = (actual baud ÷ target baud − 1) × 100 %
Baud rate calculation examples
Example 1: 9600 baud on an ATmega328P at 16 MHz
UBRR = 16,000,000 ÷ (16 × 9600) − 1 = 103.17, which rounds to 103. The real rate is 16,000,000 ÷ (16 × 104) = 9,615 baud, an error of +0.16 %. That is well inside the recommended range.
Example 2: 115200 baud on the same 16 MHz clock
At normal speed UBRR rounds to 8 and the error is -3.55 %. With U2X set, UBRR = 16 and the error is +2.12 %. Both land outside the data sheet’s recommended range (±2 % at normal speed, ±1.5 % at double speed, 8 data bits), though inside its absolute limits.
Example 3: 115200 baud on an STM32 at 72 MHz
With 16× oversampling, BRR = 72,000,000 ÷ 115,200 = 625 = 0x0271, so USARTDIV is 39.0625(mantissa 39, fraction 1/16) and the error is zero. A clock of 8 MHz gets USARTDIV 4.3125 and +0.64 %, as RM0090 Table 135 lists.
Example 4: a fractional divider on the RP2040 at 125 MHz
The PL011 UART uses 16× oversampling and a divider with 6 fractional bits. 125 MHz ÷ (16 × 115200) = 67.817, so the integer part is 67 and the fraction is 52/64, giving a divider of 67.8125, 115,207 baud and an error of +0.006 %.
UART baud rate table: UBRR at 16 MHz and 8 MHz
AVR register values and their error for standard baud rates. A dash means the rate is above what the clock reaches. For other clocks, use the AVR baud rate calculator above.
| Baud | 16 MHz, U2X = 0 | 16 MHz, U2X = 1 | 8 MHz, U2X = 0 | 8 MHz, U2X = 1 |
|---|
| 2,400 | 416 (-0.08 %) | 832 (+0.04 %) | 207 (+0.16 %) | 416 (-0.08 %) |
| 4,800 | 207 (+0.16 %) | 416 (-0.08 %) | 103 (+0.16 %) | 207 (+0.16 %) |
| 9,600 | 103 (+0.16 %) | 207 (+0.16 %) | 51 (+0.16 %) | 103 (+0.16 %) |
| 19,200 | 51 (+0.16 %) | 103 (+0.16 %) | 25 (+0.16 %) | 51 (+0.16 %) |
| 38,400 | 25 (+0.16 %) | 51 (+0.16 %) | 12 (+0.16 %) | 25 (+0.16 %) |
| 57,600 | 16 (+2.12 %) | 34 (-0.79 %) | 8 (-3.55 %) | 16 (+2.12 %) |
| 115,200 | 8 (-3.55 %) | 16 (+2.12 %) | 3 (+8.51 %) | 8 (-3.55 %) |
| 230,400 | 3 (+8.51 %) | 8 (-3.55 %) | 1 (+8.51 %) | 3 (+8.51 %) |
| 250,000 | 3 (0.00 %) | 7 (0.00 %) | 1 (0.00 %) | 3 (0.00 %) |
| 500,000 | 1 (0.00 %) | 3 (0.00 %) | 0 (0.00 %) | 1 (0.00 %) |
| 1,000,000 | 0 (0.00 %) | 1 (0.00 %) | – | 0 (0.00 %) |
STM32 BRR baud rate table, 16× oversampling
BRR and error for clocks an STM32 commonly feeds its USART. For another clock or 8× oversampling, use the STM32 baud rate calculator above.
| Baud | 72 MHz | 48 MHz | 42 MHz | 16 MHz |
|---|
| 9,600 | 0x1D4C (0.00 %) | 0x1388 (0.00 %) | 0x1117 (0.00 %) | 0x0683 (-0.02 %) |
| 19,200 | 0x0EA6 (0.00 %) | 0x09C4 (0.00 %) | 0x088C (-0.02 %) | 0x0341 (+0.04 %) |
| 38,400 | 0x0753 (0.00 %) | 0x04E2 (0.00 %) | 0x0446 (-0.02 %) | 0x01A1 (-0.08 %) |
| 57,600 | 0x04E2 (0.00 %) | 0x0341 (+0.04 %) | 0x02D9 (+0.02 %) | 0x0116 (-0.08 %) |
| 115,200 | 0x0271 (0.00 %) | 0x01A1 (-0.08 %) | 0x016D (-0.11 %) | 0x008B (-0.08 %) |
| 230,400 | 0x0139 (-0.16 %) | 0x00D0 (+0.16 %) | 0x00B6 (+0.16 %) | 0x0045 (+0.64 %) |
| 460,800 | 0x009C (+0.16 %) | 0x0068 (+0.16 %) | 0x005B (+0.16 %) | 0x0023 (-0.79 %) |
| 921,600 | 0x004E (+0.16 %) | 0x0034 (+0.16 %) | 0x002E (-0.93 %) | 0x0011 (+2.12 %) |
UART baud rate tolerance: how much error a receiver can take
A receiver lines up on the falling edge of the start bit and samples the middle of each bit after that, so the timing error builds up across the frame and is worst at the stop bit. The error budget is shared by both ends of the link.
ATmega328P, from the equations in section 20.8.3. D is data plus parity bits. Recommended values assume the transmitter and receiver split the total equally. For D = 8 at normal speed the data sheet prints −4.54 %, but its own Rslow of 95.36 % gives the −4.64 % shown here.
| D | U2X = 0 limit | Recommended | U2X = 1 limit | Recommended |
|---|
| 5 | +6.67 / −6.80 % | ±3 % | +5.66 / −5.88 % | ±2.5 % |
| 6 | +5.79 / −5.88 % | ±2.5 % | +4.92 / −5.08 % | ±2 % |
| 7 | +5.11 / −5.19 % | ±2 % | +4.35 / −4.48 % | ±1.5 % |
| 8 | +4.58 / −4.64 % | ±2 % | +3.90 / −4.00 % | ±1.5 % |
| 9 | +4.14 / −4.19 % | ±1.5 % | +3.53 / −3.61 % | ±1.5 % |
| 10 | +3.78 / −3.83 % | ±1.5 % | +3.23 / −3.30 % | ±1 % |
STM32F4 receiver tolerance from RM0090 Tables 145 and 146, with three-sample voting (ONEBIT = 0):
| Frame | Divider | 16× oversampling | 8× oversampling |
|---|
| 8-bit (M = 0) | integer (fraction 0) | ±3.75 % | ±2.5 % |
| 8-bit (M = 0) | fractional | ±3.33 % | ±2 % |
| 9-bit (M = 1) | integer (fraction 0) | ±3.41 % | ±2.27 % |
| 9-bit (M = 1) | fractional | ±3.03 % | ±1.82 % |
Those figures cover everything: transmitter error, receiver divider error, receiver clock and line asymmetry together. Maxim’s AN2141 reaches a similar general result for a 16× UART sending 8N1: ±2 % total on a long capacitive line and ±3.3 % on a short clean link.
Keeping a UART link reliable
Count the clock source in the budget, not just the divider. A 100 ppm crystal is negligible, a ceramic resonator can use up about 1 % once drift is included, and an internal RC oscillator needs checking against its data sheet. Enter these in the baud calculator’s “Other error to include” field. The divider error and the clock tolerance add, and the other device contributes its own share. If the error is too high, pick a clock that divides exactly, turn on a fractional divider, or use a lower baud rate. On STM32, make sure you use the clock feeding that USART: on STM32F4, USART1 and USART6 run from PCLK2 and the others from PCLK1.
Glossary
- Baud rate
- Symbols per second. On a UART each symbol is one bit, so this equals bits per second.
- UBRR
- The AVR USART baud rate register, 12 bits wide.
- BRR
- The STM32 USART baud rate register, holding USARTDIV as a fixed-point number.
- USARTDIV
- The STM32 baud rate divider: 12 mantissa bits and 4 (or 3) fraction bits.
- Oversampling
- How many clock ticks the receiver takes per bit to find its center, usually 16 or 8.
- U2X
- The AVR double speed bit, which changes the divider from 16 to 8.
- 8N1
- A frame of one start bit, 8 data bits, no parity and one stop bit.
Frequently asked questions
What is the UART baud rate formula?
Baud = clock ÷ (oversampling × divider). On an AVR at normal speed that is fOSC ÷ (16 × (UBRR + 1)), with 8 instead of 16 when U2X is set; on an STM32 it is fCK ÷ (8 × (2 − OVER8) × USARTDIV); and on a generic UART it is fCLK ÷ (S × D).
Baud rate vs bit rate: are they the same on a UART?
On a UART, yes. Baud rate counts symbols per second, bit rate counts bits per second, and each UART symbol is one bit, so the two numbers are equal.
What is the formula for the AVR baud rate register?
UBRR = fOSC ÷ (16 × baud) − 1, rounded to the nearest whole number. With the double speed bit U2X set, use 8 instead of 16. The actual rate is then fOSC ÷ (16 × (UBRR + 1)).
What baud rate error is acceptable for a UART?
A common rule is ±2 %. The whole link, both clocks included, typically has to stay within about ±2 % on a long or capacitive line and ±3.3 % on a short clean one. The ATmega328P data sheet recommends at most ±2 % receiver error for 8 data bits at normal speed.
What UBRR do I use for 9600 baud at 16 MHz?
UBRR = 103, which gives 9,615 baud, an error of +0.16 %. For other clocks and rates, use the UBRR calculator at the top of this page.
Why is 115200 baud unreliable on a 16 MHz Arduino?
A 16 MHz clock does not divide evenly. At normal speed, UBRR 8 gives -3.55 %; with U2X, UBRR 16 gives +2.12 %. Both are outside the data sheet’s recommended range, so the link depends on the other end being accurate.
How do I calculate the STM32 BRR value?
BRR = fCK ÷ baud with 16× oversampling, rounded to a whole number. At 72 MHz and 115200 baud that is 625 (0x0271), or USARTDIV 39.0625. With 8× oversampling the top bits are the mantissa and bits 2:0 are the fraction in eighths, and bit 3 stays clear.
What is the difference between 16× and 8× oversampling?
Dividing by 8 doubles the maximum baud rate, from fCK ÷ 16 to fCK ÷ 8, but the receiver takes half as many samples per bit, so it tolerates less clock error. RM0090 lists 3.33 % at 16× and 2 % at 8× for 8-bit frames with a fractional divider.
Which crystal frequencies give zero baud rate error?
Multiples of 1.8432 MHz, such as 3.6864, 7.3728, 11.0592, 14.7456 and 18.432 MHz, divide exactly to the standard rates. The ATmega328P data sheet tables show 0.0 % at those clocks for most rates.
More free PCB tools
See all free electronics tools →
- I2C pull-up resistor value calculatorMinimum and maximum I2C pull-up resistor for your supply voltage, bus capacitance and speed mode, per the I2C specification.
- PWM frequency calculator (STM32 and AVR)PWM frequency, resolution and duty from a timer clock, prescaler and period, or the prescaler and period for a target frequency.
- Switch debounce calculatorRC values for a hardware switch debounce circuit with a Schmitt-trigger input, from the bounce time you need to cover.
- Crystal load capacitance calculatorLoad capacitor values for a crystal oscillator from the crystal's CL and the stray capacitance, with nearest standard values.
- PPM calculator for frequency errorConvert ppm to Hz and back, clock drift in seconds per day or month, and a crystal's total frequency budget.
- 555 timer calculatorFrequency, period and duty cycle of a 555 in astable mode, or the pulse width in monostable, and the parts for a target.
Design the whole board, not one number
ProtoFlow is free AI PCB design software: describe a circuit and get a schematic with real parts, PCB layout with an autorouter, SPICE simulation and Gerber files. Native KiCad, Altium and EAGLE import and export. Free on macOS, Windows and Linux.