64bit double float to byteArray ?

Can’t you concatenate encodeNumber(n >> 32) with encodeNumber(n & 0xffffffff)?

(After zero padding them, obviously.)

See something like https://www.wikihow.com/Convert-a-Number-from-Decimal-to-IEEE-754-Floating-Point-Representation to see how doubles are encoded.

Nope, standard IEEE float representation is a bit complex with 3 parts and although 32bit and 63 bit float encoding is similar, they cannot be shifted or padded easily from one to another

Thanks. The insta360 BLE protocol does indeed use standard IEEE float representation, however I was trying to avoid burning a whole in the watch CPU,  computing 3 or 4 float values  each second ;)

So finally I've wiped up this function, not very elegant, but does the job for what I need.

	function toDouble(single) {
		// convert a 32bit float to 64-bit, unsigned, in a bitArray. 
		var double = new [0]b;
		var d1 = new [4]b;
		var d2 = new [4]b;
		var t32 = new [4]b;
		var tb = 0l;
		var tb1 = 0l;
		var tb2 = 0l;
		var te = 0l;
		var tm = 0l;
		var bias = 0l;
		// Need to convert to float 64bit, but not directly supported by Connect IQ.
		// convert back and forth from 32bit float to 32bit INT to have a "bitwise" representation
		t32.encodeNumber(single,Lang.NUMBER_FORMAT_FLOAT,null);
		tb = t32.decodeNumber(NUMBER_FORMAT_UINT32, null);
		// get the exponent part (ignoring sign)
		te = tb & 0xff800000;
		te >>= 23;
		// new 64bit based bias
		bias = te - 127 + 1023;
		bias <<= 52;
		// get the mantisse part and shift it 
		tm = tb & 0x007fffff;
		tm <<= 29;
		// add the new bias to the mantis
		tb = bias + tm;
		// cut the new value in 2 32bit chunks
		tb1 = tb;
		tb1 >>= 32;
		d1.encodeNumber(tb1, NUMBER_FORMAT_UINT32, null);
		tb1 <<=32;
		tb2 = tb - tb1;
		d2.encodeNumber(tb2, NUMBER_FORMAT_UINT32, null);
		// build the final bitArray, little endian
		double.addAll(d2);
		double.addAll(d1);
		return double;
	}

Looks good! 

If you want to improve efficiency, you may not need to calculate everything every second, but use what you had before.  For example, if speed hasn't changed, just use what you had before, or for GPS, only if it's moved more than a tiny amount (you are standing still, but may change a few feet due to GPS drift).  Elevation might not change more than a foot or 2 if you are flat and level ground, etc.

But the biggest battery drain could be BLE itself, and not the math, in which case maybe try every 5 seconds instead of every second. (or make it configurable)

Thanks for the suggestion. In any case my code needs lots of refinement ;)

And somewhat unnecessarily, going the other way:

function flBytesToType(bytes,exponentNumBits,storeBits,input) {
    	// 0. Set some constant values
    	var bias = (1 << (exponentNumBits-1)) - 1, numBytes = bytes.size();
    	var ZERO_LONG = 0l, ONE_LONG = 1l, TWO_LONG = 2l, ONE_DOUBLE = 1.0d, TWO_DOUBLE = 2.0d; // Prevent human error with wrong data type meaning shifts out of legal range...
    	// 1. Rebuild number
    	var lastByte = numBytes - 1, bitsPerByte = 8, byte = numBytes;
    	var asLong = ZERO_LONG;
    	while(byte > 0) {
    		byte--;
    		asLong |= bytes[byte].toLong() << ((lastByte - byte) * bitsPerByte);
    	}    	
    	// 2. Extract exponent and sign and mask
    	var signBit = ONE_LONG << (storeBits - 1);
    	var signMultiplier = asLong & signBit == signBit ? -1 : 1;
    	var exponent = ZERO_LONG, offset = storeBits - 1;
    	var i = 0, readPos, writePos;
    	while(i < exponentNumBits) {
    		readPos = offset - i - 1;
    		writePos = exponentNumBits - i - 1;
    		exponent |= (asLong & (ONE_LONG << readPos)) >> (readPos - writePos); 
    		i++;
    	}
    	exponent -= bias;
    	// 3. Rebuild value
		var value = ONE_DOUBLE, toAdd = ONE_DOUBLE / 2;
		var readBit = ONE_LONG << (storeBits - exponentNumBits - 2);
		while(i < storeBits) {
			if((asLong & readBit) > 0) {	value += toAdd;			}
			readBit >>= 1;
			toAdd /= 2;
			i++;
		}
		return signMultiplier * value * Math.pow(TWO_DOUBLE,exponent);    	    	
    }
    function doubleFromBytes(bytes) {
    	var input = 0d;
    	return flBytesToType(bytes,11,64,input).toDouble();
    }
    function floatFromBytes(bytes) {
    	var input = 0f;
    	return flBytesToType(bytes,8,32,input).toFloat();
    }

Thanks! You helped me a lot! Very good solution.