Selects PseudoFloat (see PseudoFloat.h) as MDFEchoCancellation's internal numeric representation, so the algorithm runs on integer mantissa/exponent arithmetic instead of native float instructions – useful on microcontrollers without an FPU. PseudoFloat's arithmetic is verified against native float directly (see the PseudoFloat tests in tests-cmake/stt/stt_test.cpp), and MDFEchoCancellation<int16_t, MDFFixedPoint> is smoke-tested to run without crashing/NaN and to converge similarly to MDFFloat on the same synthetic signal – but it has not received the same depth of numerical scrutiny as MDFFloat, so treat it as less battle-tested. Despite the name, this is not a classic narrow Q15 fixed-point format (that would over/underflow across this algorithm's actual value range – see PseudoFloat's class doc for why).
More...
#include <MDFEchoCancellation.h>
Selects PseudoFloat (see PseudoFloat.h) as MDFEchoCancellation's internal numeric representation, so the algorithm runs on integer mantissa/exponent arithmetic instead of native float instructions – useful on microcontrollers without an FPU. PseudoFloat's arithmetic is verified against native float directly (see the PseudoFloat tests in tests-cmake/stt/stt_test.cpp), and MDFEchoCancellation<int16_t, MDFFixedPoint> is smoke-tested to run without crashing/NaN and to converge similarly to MDFFloat on the same synthetic signal – but it has not received the same depth of numerical scrutiny as MDFFloat, so treat it as less battle-tested. Despite the name, this is not a classic narrow Q15 fixed-point format (that would over/underflow across this algorithm's actual value range – see PseudoFloat's class doc for why).
◆ float_t
◆ word16_t
◆ word32_t
The documentation for this struct was generated from the following file: