analysis on cost per byte of packet links

2017: So I did some figures to try to understand how expensive packet radio backbones are for the different kinds of backbone.

I used surplus prices as I believe them and with some experience playing with commercial surplus radios.

My summary is that 9600 full duplex, done with the technology I have, and assuming existing ham station Rohn 25 towers for support, not commercial sites, I decided that 9600 full duplex is quite a bit cheaper $ per byte than the lesser attempts.

However.. the total cost per station is high enough that we really need to justify NEEDing that kind of throughput. You also need to consider the other schemes. Most packet radio is done with no link planning and no consideration for the cost of collisions. Doing link planning and doing one of the earlier steps is way cheaper than doing 9600 full duplex. Please read on.

Below are my totals and my assumptions. Please feel free to send me suggestions, comments, corrections, adjustments, whatnot. What is your experience doing these kinds of things? Thanks!
Tadd, KA2DEW Raleigh NC [email protected] http://tarpn.net

I'm imagining a many hop linear network with no branches and that the backbone is tested fully loaded indefinitely. Earlier I forgot the case where people would put 9600 baud on a single-channel backbone. Added that as #2a

link types

#1. 1200 baud everybody-on-same-channel (see below) Unbelievably expensive/byte-per-second. not worth doing.
#2. 1200 baud private-single-frequency long multihop backbone $105/byte-per-second
#2a 9600 baud private-single-frequency long multihop backbone $58/byte-per-second
#3. 1200 baud point-to-point simplex links where each site to site is own pair of radios and own frequency $13/byte-per-second
#4. 9600 baud point-to-point simplex links where each site to site is own pair of radios and own frequency $10/byte-per-second
#5. 1200 baud full-duplex links $17/byte-per-second
#6. 9600 baud full-duplex links. $3/byte-per-second

The packet organization that I've been working with, TARPN, is advocating case #3 because it does NOT take a knowledge of heliax, duplexers and precision tuning of levels and whatnot to build. It also has a low barrier to entry compared to options #2a, 4, 5, and 6. Our total test equipment is a dedicated $20 oscope kit and a as-needed handie talkie or portable receiver on each band. We assume a 3 port node costs about $800.

I assume that the backbones we are discussing will be fully loaded so the numbers I'm tossing around assume that loading though I give a little grace for the many-stations-on-channel scenarios because otherwise they would have 0 performance.
I'm assuming 1200 done with TNC-PI @$50 and Raspberry PI with G8BPQ for simplex and some used TNC2@$100 and Raspberry PI for duplex.
I'm assuming 9600 done with TimeWave PK-96 @$200 and Raspberry PI with G8BPQ -- I don't know how much the TNC-PI9k6 costs. Will revise later. DireWolf would be cheaper but will require some physical engineering for the sound cards, PTT, timeouts, and some failsafe restart system. Not ready yet. Not important. Read on.
For the purposes of discussion I presume a dedicated link quiet channel with two radios with PPERSIST turned completely to full and 1200 baud modems with normal TXDELAY we get about 40 characters per second.
For 9600, same setup we get about 120 characters per second with a dedicated link.
The trick with all-backbone-on-single-channel packet networks is that they work pretty well until the channel saturation reaches 20% and then they work horribly! But since the only solution to keep the channel saturation to below 20% is Ppersistance or "backoff" the 20% figure is bumped up against over and over again.
For a simplex all-backbone-on-single-channel i.e. no channel access control, 20% channel loading is available before nearly continuous collisions start, I presume a 90% performance hit over simplex dedicated link. Sometimes it will be worse! So 4chars/second @1200 and 12chars/second @ 9600. From what I can tell, with 9600 baud, the levels at adjacent stations are critical, so a multi-station-on-frequency 9600 backbone is tough to align.
For a full duplex link I think can get on the order of 100 characters per second total of both directions at 1200 baud.
For a full duplex link I think can get on the order of 800 characters per second total of both directions at 9600 baud.
Assume each site is at a ham's existing (free) Rohn triangular tower where the top is NOT available. Commercial sites will be more difficult especially for full duplex because the installation would have imposed limitations due to other existing transmitters and receivers. Also, the antennas have to be falling-ice-proof and wont be serviced readily after a storm. More expensive antennas will be needed.
Assume 100' run between antennas and radios using Jefatech LMR400 clone at $0.60 cents per foot. $60 per run. For full duplex we need Heliax, surplus at $2/foot + $40 surplus connectors = $240 per run.
These are my expectation of site equipment to do the 'backbone'. This is irrespective of other linking equipment and not taking into account interference with other equipment at the site.
I assume full duplex is same-band for each direction, but different bands for north vs south.
For full duplex links I presume that the $70 commercial surplus radio works ok with a little extra work, fans and whatnot? Let's add a $100 per transmitter tax?
Also for full duplex links I presume we need a set of pass-can duplexers, surplus for $200 per link, 2 per site.
For 9600 baud I make a wild guess that a modified commercial radio can be made to work half for half the links?
For the other half we'll go with a ham cost radio at $300 for 220 or 440Mhz. I'm a pessimist about getting radios to work at 9600 but feel free to help if I'm wrong.
Let's hedge my pessimism and say $200 per radio for 9600?
I've seen commercial radios work at 9600 but not recently and the radios used were not very common.
Please give me feedback on this. Can we do this for under $200 per radio? Where? What do the eye patterns look like? I need help here.
Kantronics D4-10 radios are a good deal at $400 per end. I have been watching eBay for a year waiting for more to come by. I have two Kantronics D4-10 radios and for ham site linking they are amazing but only 10 watts and only on 440. Not sure what they would be like in a high RF environment. If you have any of these to sell Id like to talk.
Omni antennas I will say cost $200 for a trivial side-arm mounted gain omni. Vertical dipoles would be cheaper but are also expensive if set up for any amount of gain. For Yagis I presume about $70 per end-mounted antenna.

#1 1200 baud everybody-on-same-channel Not a backbone.

not interesting. Like #2 but assume no better than 2 bytes per second?? Depends on the neighborhood? Out in the country in western NC, they tell me this works at better than 20 characters per second but only if there is only one station on-the-air at a time. If you are thinking about full duplex, then this isnt where you live. My experience is that this crashes frequently because the relay sites can all see each other so they get hung up waiting for the next one down the line to stop transmitting while the ground-stations keep pounding with retries. Yuck
See "TARPN Networking On Purpose"

#2 1200 baud private-single-frequency long multihop backbone

One radio for the backbone, likely to be on 6m, 220 or 440 band. Any FM rig will do. 30% estimated duty cycle. $70
One coax run, $60.
One omni directional antenna. Side-arm for tower required. $200 for the antenna
One TNC-PI = $50
Raspberry pI = $40
Total equipment cost
Throughput of 4 characters per second
Total cost W.A.G. is $420
Cost of bytes per second is on the order of $105 per byte per second.

#2a 9600 baud private-single-frequency long multihop backboneone radio for the backbone.

Likely to be on 220 or 440 band. $200 radio required. 30% estimated duty cycle $200 One coax run, $60.
One omni directional antenna. Side-arm for tower required. $200 for the antenna
One PK-96 = $200
Raspberry pI = $40
Throughput of 12 characters per second
Total cost W.A.G. is $700
Cost of bytes per second is on the order of $58 per byte per second.

#3 1200 baud point-to-point simplex links where each site to site is own pair of radios and own frequency

Two radios for backbone. Likely to be on 6m, 220 or 440band. Any FM rig will do. 50% estimated duty cycle $70 x 2 = $140.
Two coax runs, or one coax run and diplexers to cross-band. $60 * 2 = $120
Two yagi antennas. No side-arm required for tower mount. $70 per antenna * 2 = $140
Two TNC-PI TNCs + Raspberry PI $120
Total cost W.A.G. $520
Throughput of 40 bytes per second reliably.
Cost of bytes per second is on the order of $13 per byte per second

#4 9600 baud point-to-point simplex links where each site to site is own pair of radios and own frequency

Same as #3 but with more expensive radios.
Throughput is about 120 bytes per second because the rest of the bytes per second are eaten by key-up delays.
Two high speed capable radios for backbone. 220 or 440band. 50% estimated duty cycle $200 x 2 = $400.
Two coax runs, or one coax run and diplexers to cross-band. $60 * 2 = $120
Two yagi antennas. No side-arm required for tower mount. $70 per antenna * 2 = $140
Two PK-96 + Raspberry PI = $200 + $200 + $40 = $480 (This could subtract $300+ if using emulated TNC!)
Total cost W.A.G. $1140
Throughput of 120 bytes per second reliably. Cost of bytes per second is on the order of $10 per byte per second

#5 1200 baud full-duplex links

Throughput is about 120 bytes per second because the rest of the bytes per second are eaten by key-up delays.
Four radios for backbone. 220 or 440band. 100% estimated duty cycle -- add $100 per transmitter ($70 x 4) + ($100 x 2) = $480.
Two Heliax runs. $240 * 2 = $480
Two pass can duplexers, surplus $200 * 2 = $400
Two yagi antennas. No side-arm required for tower mount. $70 per antenna * 2 = $140
Two TNC2 + Raspberry PI. $100 + $100 + $40 = $240 (This could subtract $100+ if using emulated TNC!)
Total cost W.A.G. $1740
Throughput of 100 bytes per second reliably. Cost of bytes per second is on the order of $17 per byte per second

#6 9600 baud full-duplex links.

Throughput is about 800 bytes per second because the rest of the bytes per second are eaten by key-up delays.
Four radios for backbone. 220 or 440band. 100% estimated duty cycle -- add $100 per transmitter ($200 x 4) + ($100 x 2) = $1000.
Two Heliax runs. $240 * 2 = $480
Two pass can duplexers, surplus $200 * 2 = $400
Two yagi antennas. No side-arm required for tower mount. $70 per antenna * 2 = $140
Two PK96 + Raspberry PI. $200 + $200 + $40 = $440 (This could subtract $300+ if using emulated TNC!)
Total cost W.A.G. $2460
Throughput of 800 bytes per second reliably.
Cost of bytes per second is on the order of $3 per byte per second