Magnetic resonance image quality, e.g., SNR, resolution and contrast, benefits greatly with the increased static magnetic field strength B0 [1]. However, RF power deposition and specific absorption rate (SAR) also increase near quadratically with B0 [2]. Therefore, SAR becomes a limiting factor in high field MRI for various applications. Multi-channel RF transmission also results in electromagnetic field variations that may cause the local peak SAR to exceed the IEC/FDA limits [3, 4]. Under these conditions, monitoring the RF power deposition and global and local peak SAR values are of vital significance to patient safety. In this study, we developed a multi-channel real time RF power monitoring system for SAR estimation using the FPGA technology in high field MRI.Eight forward and reflected RF power monitoring pathways were built to detect the transmit RF powers delivered to the MRI coil arrays. The overall block diagram is shown in Fig. 1. The power monitoring pathways include: dual directional couplers, power sensor modules, analog-to-digital converters (ADC) and the FPGA unit. In each transmission pathway, the forward and reflected RF powers were measured by a non-magnetic dual directional coupler (Pulsar Microwave, NJ, USA ) followed with two AD8309 (Analog Devices, Norwood, MA) demodulating logarithmic power sensors [5]. The directional coupler can couple the peak RF power up to 2000w with the coupling factor of 50 dB. The AD8309 further demodulated the RF power signal with a dynamic range between -78dBm to 22dBm. The power monitoring pathway therefore can detect the peak input RF power varying from 0.0016mW up to 15kW. The DC signals from 16 power sensors were then digitized by two 12-bit, 8-channel AD7298 (Analog Devices, Norwood MA) ADCs, and the data stream were read by a Xilinx fast FPGA signal processing unit. FPGA processed the multi-channel digitized signals in a high speed, time-sharing mode. The deposited RF power was then calibrated for system errors with a loop-up table and continuously averaged for a short 10-second and a long 6-minute period. The averaged RF power values were then transferred to a PC computer for further global and local peak SAR estimation with algorithms. The power absorption ratio to estimate the global and local peak SAR values over 10 gram tissue in a human subject model was determined from the EM simulation results, while the SAR threshold values were decided according to the FDA and IEC criteria [4]. The PC computer would send a TTL triggering signal to the MRI spectrometer when the average power of global or local peak SAR exceeded the limits.A multi-layer PCB board with the integration of the16-channel AD8309 power sensors, two AD7298 high-speed ADC chips and the FPGA unit was built and shown in Fig. 2. To test the function of the RF power monitoring system, an Agilent N5181A RF signal generator was employed to produce a 300MHz sinusoidal RF signal, which was then equally split into 8-channels and fed into the eight forward RF power monitoring pathways. The measurement for each channel was calibrated individually at the operational frequency (300MHz) in comparison with an Agilent 8990B peak power analyzer. The maximal error of the measurements was less than 0.5dB for the input RF power ranged from -45dBm to 15 dBm, corresponding to a measurement error of less than 11%. An example of measured forward, reverse and averaged RF power deposition over 10 seconds and 6 minutes for individual and combined RF channels, with the corresponding estimated global SAR and local peak SAR values over 10 gram tissue of an human subject model, were displayed in a home-developed, user-friendly Qt5.3.2 (Nokia, Espoo, Finland) application program in the aforementioned PC computer (Fig. 3). The communication between PC and FPGA is based on a high-speed user datagram protocol (UDP) which can transfer the data at the rate of 50 Mbps. We have developed an 8-channel FPGA-based real-time power monitoring system for global and local SAR estimation to ensure patient safety for high field MRI. The monitoring pathway exhibited a large dynamic range of RF signal for about 100dB. The total insertion loss of the monitoring system was about 0.2dB, which comes mainly from the directional coupler. The error of the RF power measurement was within 0.5dB after calibration. The high speed FPGA unit ensured fast processing of forward and reverse RF power averaged over 10 seconds and 6 minutes separately for each channel. The global and local peak SAR values were estimated with accuracy based on RF power measurements and EM simulation results of a human subject model.