If you data will fit into an advertising packet payload, then that is the lowest power possible way to get that data delivered to a receiver (or receivers). To send an atomic data set via advertising, you only need to send a single packet. A connection at very least would require a send followed by a receive.
You also indicate that you could go hours between data packets. With advertising, you can send your packets spontaneously as you like- as long as the receiver gets the packet when you do send it then the transfer is good.
Advertising packets are also much simpler to create. You could save processor cycles by laying out all the static parts of the packet in memory one and then very quickly updating only the dynamic data fields before a send.
You need to do what we call a link budget. This typically takes into account:
- the directivity of the antenna/its gain with respect to an omnidirectional antenna (transmitter and receiver)
- the power converted into waves in the antenna. Not to mix up with the power in the transmitter, the efficiency of the antenna driver can vary quite a bit from transmitter to transmitter!
- The free space loss which depends on frequency of the signal (related to baudrate) and distance (absorption of air)
Such a link budget sums all gains and losses and always equals zero, so you can solve for the maximum baudrate, knowing that the signal power at the receiver must be X times bigger than the estimated noise power at the receiver.
X is the signal to noise ratio, which is related to the bit error rate via the modulation technique that is used by the transmitter/receiver. i.e. Certain modulation techniques are more robust against noise than others and don't generate that many upsets.
Usually there is something else on top of the hardware layer which is encoding, to decrease further the bit error rate for a given modulation technique. In its simplest form, it's a parity bit.
With those keywords, you should be able to know what the specs are for bluetooth and calculate the maximum baudrate.
Or you could do it "the experimental way", and run the system for several days at a constant baudrate and record the number of bit upsets. If it's fine to you, stick to it. The above is only for your own understanding, if you can actually test it, test it.
Best Answer
The goal of this rather long analysis is to ensure you look like the sharpest RF dude your company has ever hired :)
The free space path loss range of a an RF system can be calculated.
For a typical class 2 Bluetooth system, assuming +4dBm of transmit power, -70dBm receiver, & unity gain antennas, you will find that the range that gives you 74 dB of path loss (the maximum path loss that still works) is 50 meters. Below I did it in one of the many online path loss calculators:
From the path loss equation you will note that power is a function of the square of range, so, if you reduce the range by a factor of 2, the required transmit power (or receive sensitivity) falls by a factor of 4. (i.e. 6dB).
For example, if we reduce the transmit power from 4 dB to -2dB, the free space range will reduce to 25 meters. Lose another 6dB and you are around 12.5 meters.
Sounds easy huh? Nope...
Mobile links such as Bluetooth move around in a complex environment of objects that randomly reflect the signals. At the receiver, these multiple paths can either support or cancel each other, so as the mobile unit moves around, the signal power fluctuates wildly. How wildly? Well, it turns out that it's all statistics: The received signal strength turns out to be our old friend the normal distribution. The average signal strength is simply what is predicted by the path loss equation.
So first bad news is that if you depended on that 50m free space range, your link will only work half of the time. (top half of the distribution)
The second bad news is the standard deviation, which in this case turns out to be roughly 6dB. You'll see from this rule that 99.7% of signal strengths fall within -3 and +3 standard deviations.
That can be stated in a different way, and you are not going to like it: Your mobile Bluetooth signal will vary over a -18dB to +18db range around the mean value predicted by the path loss model!
Since range-squared and power is related through the first equation, we can also state the rule-of-thumb that your operating range will regularly vary between (roughly) 1/8x to 8x of the 50m estimated free space value!
In practice, this means that for around 0.15% of the time, you won't even be able to make the 50m link work at 50/8 = 6 meters, but on the other hand, for .15% of the time you will "get lucky" and the link will work as far out as 400 meters. If you are a smart product marketer, you will quote the range as 6 meters. Your product will meet your spec 99.85% of the time!
Why this long story? To provide you with sound reasoning regarding why you have to be very alert to the statistical nature of mobile links like Blueooth!
Your intended RF minimum distance is going going to vary over a 1:64 range! Is that OK?
If the range you are thinking of is less than, say, one meter, inductively coupled systems could be your answer, the key reason being that unlike RF, where received power is a function of 1/range-squared, in magnetically coupled systems power drops with 1/range-cubed. That extra order really takes care of "accidental" successes at further distances, but it's also the reason why the range is typically constrained to less than a meter.
Also, you might want to look at Alan Bensky's book on Wireless Positioning Technologies and Applications.