The accuracy of the bq27621 fuel gauge is disappointing. In particular the state-of-charge (SOC) indication appears substantially too low when the battery is heavily loaded. This indication falls below 3% after about 62 minutes of full operation with the Colibri T20 (Computer on Module), and the power turns off shortly thereafter as a result of the 2% SOCF threshold. However, if we then interrogate the gauge's SOC with TI’s Battery Management System, it shows that removing the load has revised the SOC to 28%. If this value had been measured under load, it would imply a time to shut-off of about 86 minutes, which is typical for the discharge time for the current firmware. Thus, the loaded SOC has large enough errors such that it cannot be used to turn off power unless a significant amount of battery capacity is given up. The problem does not lie with the PIC firmware but with the gauge's operation.
We already tried two things to attempt to resolve this problem. First, we made a hard connection from the battery to the gauge inputs to reduce a voltage drop in this path from 25 mV to 4 mV. This change appeared to make the SOC decay with time more uniform but did not have a major effect on the measured SOC. Then we tried changing the Load Select/Mode parameter from 0x80 to 0x83 to change the Current Model from the average power for the previous cycle, which appeared much too low, to the instantaneous power after a 14-s low-pass filter. These changes resulted in the above-described performance, which is not substantially different from that before the changes.
We’re not sure where we go from here. Does anyone have any advice? Possibly some other obscure initialization parameter could fix the problem, but we cannot find it. Or maybe our battery chemistry is simply not the same as that for the standard LiCoO2-based batteries for which the gauge was designed, even though our maximum charging voltage is 4.2 V.