/* This project is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. Multiprotocol is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with Multiprotocol. If not, see . */ //--------------------------- // AVR nrf chip bitbang SPI functions //--------------------------- #include "iface_nrf24l01.h" //--------------------------- // NRF24L01+ SPI Specific Functions //--------------------------- uint8_t rf_setup; void NRF24L01_Initialize() { rf_setup = 0x09; } void NRF24L01_WriteReg(uint8_t reg, uint8_t data) { NRF_CSN_off; SPI_Write(W_REGISTER | (REGISTER_MASK & reg)); SPI_Write(data); NRF_CSN_on; } void NRF24L01_WriteRegisterMulti(uint8_t reg, uint8_t * data, uint8_t length) { NRF_CSN_off; SPI_Write(W_REGISTER | ( REGISTER_MASK & reg)); for (uint8_t i = 0; i < length; i++) SPI_Write(data[i]); NRF_CSN_on; } void NRF24L01_WritePayload(uint8_t * data, uint8_t length) { NRF_CSN_off; SPI_Write(W_TX_PAYLOAD); for (uint8_t i = 0; i < length; i++) SPI_Write(data[i]); NRF_CSN_on; } uint8_t NRF24L01_ReadReg(uint8_t reg) { NRF_CSN_off; SPI_Write(R_REGISTER | (REGISTER_MASK & reg)); uint8_t data = SPI_Read(); NRF_CSN_on; return data; } /*static void NRF24L01_ReadRegisterMulti(uint8_t reg, uint8_t * data, uint8_t length) { NRF_CSN_off; SPI_Write(R_REGISTER | (REGISTER_MASK & reg)); for(uint8_t i = 0; i < length; i++) data[i] = SPI_Read(); NRF_CSN_on; } */ static uint8_t __attribute__((unused)) NRF24L01_ReadPayloadLength() { NRF_CSN_off; SPI_Write(R_RX_PL_WID); uint8_t len = SPI_Read(); NRF_CSN_on; return len; } static void NRF24L01_ReadPayload(uint8_t * data, uint8_t length) { NRF_CSN_off; SPI_Write(R_RX_PAYLOAD); for(uint8_t i = 0; i < length; i++) data[i] = SPI_Read(); NRF_CSN_on; } static void NRF24L01_Strobe(uint8_t state) { NRF_CSN_off; SPI_Write(state); NRF_CSN_on; } void NRF24L01_FlushTx() { NRF24L01_Strobe(FLUSH_TX); } void NRF24L01_FlushRx() { NRF24L01_Strobe(FLUSH_RX); } void NRF24L01_Activate(uint8_t code) { NRF_CSN_off; SPI_Write(ACTIVATE); SPI_Write(code); NRF_CSN_on; } void NRF24L01_SetBitrate(uint8_t bitrate) { // Note that bitrate 250kbps (and bit RF_DR_LOW) is valid only // for nRF24L01+. There is no way to programmatically tell it from // older version, nRF24L01, but the older is practically phased out // by Nordic, so we assume that we deal with modern version. // Bit 0 goes to RF_DR_HIGH, bit 1 - to RF_DR_LOW rf_setup = (rf_setup & 0xD7) | ((bitrate & 0x02) << 4) | ((bitrate & 0x01) << 3); NRF24L01_WriteReg(NRF24L01_06_RF_SETUP, rf_setup); } /* static void NRF24L01_SetPower_Value(uint8_t power) { uint8_t nrf_power = 0; switch(power) { case TXPOWER_100uW: nrf_power = 0; break; case TXPOWER_300uW: nrf_power = 0; break; case TXPOWER_1mW: nrf_power = 0; break; case TXPOWER_3mW: nrf_power = 1; break; case TXPOWER_10mW: nrf_power = 1; break; case TXPOWER_30mW: nrf_power = 2; break; case TXPOWER_100mW: nrf_power = 3; break; case TXPOWER_150mW: nrf_power = 3; break; default: nrf_power = 0; break; }; // Power is in range 0..3 for nRF24L01 rf_setup = (rf_setup & 0xF9) | ((nrf_power & 0x03) << 1); NRF24L01_WriteReg(NRF24L01_06_RF_SETUP, rf_setup); } */ void NRF24L01_SetPower() { uint8_t power=NRF_BIND_POWER; if(IS_BIND_DONE_on) power=IS_POWER_FLAG_on?NRF_HIGH_POWER:NRF_LOW_POWER; if(IS_RANGE_FLAG_on) power=NRF_POWER_0; rf_setup = (rf_setup & 0xF9) | (power << 1); if(prev_power != power) { NRF24L01_WriteReg(NRF24L01_06_RF_SETUP, rf_setup); prev_power=power; } } void NRF24L01_SetTxRxMode(enum TXRX_State mode) { if(mode == TX_EN) { NRF_CE_off; NRF24L01_WriteReg(NRF24L01_07_STATUS, (1 << NRF24L01_07_RX_DR) //reset the flag(s) | (1 << NRF24L01_07_TX_DS) | (1 << NRF24L01_07_MAX_RT)); NRF24L01_WriteReg(NRF24L01_00_CONFIG, (1 << NRF24L01_00_EN_CRC) // switch to TX mode | (1 << NRF24L01_00_CRCO) | (1 << NRF24L01_00_PWR_UP)); delayMicroseconds(130); NRF_CE_on; } else if (mode == RX_EN) { NRF_CE_off; NRF24L01_WriteReg(NRF24L01_07_STATUS, 0x70); // reset the flag(s) NRF24L01_WriteReg(NRF24L01_00_CONFIG, 0x0F); // switch to RX mode NRF24L01_WriteReg(NRF24L01_07_STATUS, (1 << NRF24L01_07_RX_DR) //reset the flag(s) | (1 << NRF24L01_07_TX_DS) | (1 << NRF24L01_07_MAX_RT)); NRF24L01_WriteReg(NRF24L01_00_CONFIG, (1 << NRF24L01_00_EN_CRC) // switch to RX mode | (1 << NRF24L01_00_CRCO) | (1 << NRF24L01_00_PWR_UP) | (1 << NRF24L01_00_PRIM_RX)); delayMicroseconds(130); NRF_CE_on; } else { NRF24L01_WriteReg(NRF24L01_00_CONFIG, (1 << NRF24L01_00_EN_CRC)); //PowerDown NRF_CE_off; } } void NRF24L01_Reset() { //** not in deviation but needed to hot switch between models NRF24L01_Activate(0x73); // Activate feature register NRF24L01_WriteReg(NRF24L01_1C_DYNPD, 0x00); // Disable dynamic payload length on all pipes NRF24L01_WriteReg(NRF24L01_1D_FEATURE, 0x00); // Set feature bits off NRF24L01_Activate(0x73); //** NRF24L01_FlushTx(); NRF24L01_FlushRx(); NRF24L01_Strobe(0xff); // NOP NRF24L01_ReadReg(NRF24L01_07_STATUS); NRF24L01_SetTxRxMode(TXRX_OFF); delayMicroseconds(100); } uint8_t NRF24L01_packet_ack() { switch (NRF24L01_ReadReg(NRF24L01_07_STATUS) & (BV(NRF24L01_07_TX_DS) | BV(NRF24L01_07_MAX_RT))) { case BV(NRF24L01_07_TX_DS): return PKT_ACKED; case BV(NRF24L01_07_MAX_RT): return PKT_TIMEOUT; } return PKT_PENDING; } /////////////// // XN297 emulation layer uint8_t xn297_scramble_enabled=XN297_SCRAMBLED; //enabled by default uint8_t xn297_addr_len; uint8_t xn297_tx_addr[5]; uint8_t xn297_rx_addr[5]; uint8_t xn297_crc = 0; static const uint8_t xn297_scramble[] = { 0xe3, 0xb1, 0x4b, 0xea, 0x85, 0xbc, 0xe5, 0x66, 0x0d, 0xae, 0x8c, 0x88, 0x12, 0x69, 0xee, 0x1f, 0xc7, 0x62, 0x97, 0xd5, 0x0b, 0x79, 0xca, 0xcc, 0x1b, 0x5d, 0x19, 0x10, 0x24, 0xd3, 0xdc, 0x3f, 0x8e, 0xc5, 0x2f}; const uint16_t PROGMEM xn297_crc_xorout_scrambled[] = { 0x0000, 0x3448, 0x9BA7, 0x8BBB, 0x85E1, 0x3E8C, 0x451E, 0x18E6, 0x6B24, 0xE7AB, 0x3828, 0x814B, 0xD461, 0xF494, 0x2503, 0x691D, 0xFE8B, 0x9BA7, 0x8B17, 0x2920, 0x8B5F, 0x61B1, 0xD391, 0x7401, 0x2138, 0x129F, 0xB3A0, 0x2988}; const uint16_t PROGMEM xn297_crc_xorout[] = { 0x0000, 0x3d5f, 0xa6f1, 0x3a23, 0xaa16, 0x1caf, 0x62b2, 0xe0eb, 0x0821, 0xbe07, 0x5f1a, 0xaf15, 0x4f0a, 0xad24, 0x5e48, 0xed34, 0x068c, 0xf2c9, 0x1852, 0xdf36, 0x129d, 0xb17c, 0xd5f5, 0x70d7, 0xb798, 0x5133, 0x67db, 0xd94e}; static uint8_t bit_reverse(uint8_t b_in) { uint8_t b_out = 0; for (uint8_t i = 0; i < 8; ++i) { b_out = (b_out << 1) | (b_in & 1); b_in >>= 1; } return b_out; } static const uint16_t polynomial = 0x1021; static uint16_t crc16_update(uint16_t crc, uint8_t a) { crc ^= a << 8; for (uint8_t i = 0; i < 8; ++i) if (crc & 0x8000) crc = (crc << 1) ^ polynomial; else crc = crc << 1; return crc; } void XN297_SetTXAddr(const uint8_t* addr, uint8_t len) { if (len > 5) len = 5; if (len < 3) len = 3; uint8_t buf[] = { 0x55, 0x0F, 0x71, 0x0C, 0x00 }; // bytes for XN297 preamble 0xC710F55 (28 bit) xn297_addr_len = len; if (xn297_addr_len < 4) for (uint8_t i = 0; i < 4; ++i) buf[i] = buf[i+1]; NRF24L01_WriteReg(NRF24L01_03_SETUP_AW, len-2); NRF24L01_WriteRegisterMulti(NRF24L01_10_TX_ADDR, buf, 5); // Receive address is complicated. We need to use scrambled actual address as a receive address // but the TX code now assumes fixed 4-byte transmit address for preamble. We need to adjust it // first. Also, if the scrambled address begins with 1 nRF24 will look for preamble byte 0xAA // instead of 0x55 to ensure enough 0-1 transitions to tune the receiver. Still need to experiment // with receiving signals. memcpy(xn297_tx_addr, addr, len); } void XN297_SetRXAddr(const uint8_t* addr, uint8_t len) { if (len > 5) len = 5; if (len < 3) len = 3; uint8_t buf[] = { 0, 0, 0, 0, 0 }; memcpy(buf, addr, len); memcpy(xn297_rx_addr, addr, len); for (uint8_t i = 0; i < xn297_addr_len; ++i) { buf[i] = xn297_rx_addr[i]; if(xn297_scramble_enabled) buf[i] ^= xn297_scramble[xn297_addr_len-i-1]; } NRF24L01_WriteReg(NRF24L01_03_SETUP_AW, len-2); NRF24L01_WriteRegisterMulti(NRF24L01_0A_RX_ADDR_P0, buf, 5); } void XN297_Configure(uint8_t flags) { xn297_crc = !!(flags & BV(NRF24L01_00_EN_CRC)); flags &= ~(BV(NRF24L01_00_EN_CRC) | BV(NRF24L01_00_CRCO)); NRF24L01_WriteReg(NRF24L01_00_CONFIG, flags & 0xFF); } void XN297_SetScrambledMode(const u8 mode) { xn297_scramble_enabled = mode; } void XN297_WritePayload(uint8_t* msg, uint8_t len) { uint8_t buf[32]; uint8_t last = 0; if (xn297_addr_len < 4) { // If address length (which is defined by receive address length) // is less than 4 the TX address can't fit the preamble, so the last // byte goes here buf[last++] = 0x55; } for (uint8_t i = 0; i < xn297_addr_len; ++i) { buf[last] = xn297_tx_addr[xn297_addr_len-i-1]; if(xn297_scramble_enabled) buf[last] ^= xn297_scramble[i]; last++; } for (uint8_t i = 0; i < len; ++i) { // bit-reverse bytes in packet uint8_t b_out = bit_reverse(msg[i]); buf[last] = b_out; if(xn297_scramble_enabled) buf[last] ^= xn297_scramble[xn297_addr_len+i]; last++; } if (xn297_crc) { uint8_t offset = xn297_addr_len < 4 ? 1 : 0; uint16_t crc = 0xb5d2; for (uint8_t i = offset; i < last; ++i) crc = crc16_update(crc, buf[i]); if(xn297_scramble_enabled) crc ^= pgm_read_word(&xn297_crc_xorout_scrambled[xn297_addr_len - 3 + len]); else crc ^= pgm_read_word(&xn297_crc_xorout[xn297_addr_len - 3 + len]); buf[last++] = crc >> 8; buf[last++] = crc & 0xff; } NRF24L01_WritePayload(buf, last); } void XN297_ReadPayload(uint8_t* msg, uint8_t len) { // TODO: if xn297_crc==1, check CRC before filling *msg NRF24L01_ReadPayload(msg, len); for(uint8_t i=0; i5?5:pos; } void LT8900_SetAddress(uint8_t *address,uint8_t addr_size) { uint8_t addr[5]; //Address size (SyncWord) 2 to 8 bytes, 16/32/48/64 bits LT8900_addr_size=addr_size; for (uint8_t i = 0; i < addr_size; i++) LT8900_addr[i] = address[addr_size-1-i]; //Build overhead LT8900_BuildOverhead(); //Set NRF RX&TX address based on overhead content NRF24L01_WriteReg(NRF24L01_03_SETUP_AW, LT8900_buffer_start-2); for(uint8_t i=0;i>8)&0xFF; } //Check len if(LT8900_Flags&_BV(LT8900_PACKET_LENGTH_EN)) { crc=crc16_update(crc,buffer[pos]); if(bit_reverse(len)!=buffer[pos++]) return 0; // wrong len... } //Decode message for(i=0;i>8)&0xFF)) return 0; // wrong CRC... if(buffer[pos]!=(crc&0xFF)) return 0; // wrong CRC... } //Everything ok return 1; } void LT8900_WritePayload(uint8_t* msg, uint8_t len) { unsigned int crc=LT8900_CRC_Initial_Data,a,mask; uint8_t i, pos=0,tmp, buffer[64], pos_final,shift; //Add packet len if(LT8900_Flags&_BV(LT8900_PACKET_LENGTH_EN)) { tmp=bit_reverse(len); buffer[pos++]=tmp; crc=crc16_update(crc,tmp); } //Add payload for(i=0;i>8; buffer[pos++]=crc; } //Shift everything to fit behind the trailer (4 to 18 bits) shift=LT8900_buffer_overhead_bits&0x7; pos_final=LT8900_buffer_overhead_bits/8; mask=~(0xFF<<(8-shift)); LT8900_buffer[pos_final+pos]=0xFF; for(i=pos-1;i!=0xFF;i--) { a=buffer[i]<<(8-shift); LT8900_buffer[pos_final+i]=(LT8900_buffer[pos_final+i]&mask>>8)|a>>8; LT8900_buffer[pos_final+i+1]=(LT8900_buffer[pos_final+i+1]&mask)|a; } if(shift) pos++; //Send everything NRF24L01_WritePayload(LT8900_buffer+LT8900_buffer_start,pos_final+pos-LT8900_buffer_start); } // End of LT8900 emulation