This is a companion project to my small transmitting loop experiments. Those experiments have produced a number of loop antennas of different sizes and configurations, and have demonstrated to my satisfaction that the small transmitting loop can be a serious HF antenna system.

Designing such an antenna involves balancing a number of variables, such as overall size, shape, conductor diameter and material, capacitor voltage and current ratings, desired power handling, and of course the desired frequency range. Some of these variables are critical in the sense that small changes in their values can lead to large changes in the performance of the antenna for any given frequency. As such, any such antenna is a compromise between these variables. When a loop is intended for use on more than one band, it can be difficult to find the right combination of measurements that will optimize the antenna such that it is the best compromise across the intended bands of operation.

These design challenges prompted me to look for ways to automate the design process, so that I could select which characteristics are important to me, and allow a computer to determine the details to make the best overall antenna for the constraints I provide. This has led me to experiment with a tool I first encountered in graduate school --

The Wikipedia article linked above provides a better explanation of the so-called "genetic algorithms" than I could repeat here. I would summarize these tools as a way to do

In this context, I am using a GA to generate and test different loop antenna designs, to optimize a single antenna for use on one or more bands. In this way, I don't need to formulate the best antenna parameters, but I

Where this approach really shines is when solving problems where it is impossible or impractical to devise a system of equations to solve for an optimal solution, but where there is a measure for comparing different designs to determine which is the closest to the desired optimal solution. In these types of problems, the GA can be essentially another form of numerical computing, determining a solution that is "good enough" for the needs of the user, without requiring a rigorous mathematical approach to the solution.

The core of any GA is the fitness function. This function is used to measure each proposed design generated by the GA, and is used to filter or "weed out" the less useful designs in favor of those with better characteristics. In the first draft of these experiments, I have chosen a very simple fitness function, which is a product of four even simpler functions, evaluated together for a specific target frequency,

This function tries to maximize all four of these items in a resulting antenna design, and it treats all of them as roughly equal in priority.

The GA generates hundreds or thousands of antenna designs, and uses this overall function to evaluate each design for each of the bands or frequencies I request. The GA proceeds through several "generations" of designs, and between each generation, it attempts to improve the next generation of designs based on the results of the current generation. In the end, if the GA is configured properly, the final generation should consist of several designs that are all close to convergence on a single set of design numbers that I could use to build the best-fit antenna for the constraints I have provided.

The fitness function described above has produced several reasonable antenna designs. The quality of the results depends on the constraints provided, as with any GA. But here are a couple of examples that were computed using a generation size of 100 and a limit of 65 generations, at which time the GA had largely converged to a single preferred solution. For those of you fluent in GA/GP/EC techniques, the genome was a floating-point vector of size two, and each of the floating-point numbers in the vector represents a specific measurement in the antenna design; the first is the overall loop diameter, measured in feet, and the second is the loop conductor size, measured in inches. Each individual generated is represented by this pair of numbers.

The algorithm generated a 40m-only circular copper loop that was optimized for that band as described above. The loop diameter recommended by the GA was

According to the equations taken from Chapter 5 of the ARRL Antenna Book, this antenna will have an overall efficiency of 85.5% (less than 1dB of conductor loss), a +/- 3dB bandwidth of 12.7kHz, and when driven with a 100W CW transmitter it will generate a working capacitor voltage of 3.9kV and a circulating current of 14.5A. As loop designs go, that one is quite good. The loop is just under 8' tall, but can be run very comfortably at 100W with a capacitor rated at only 5kV, and it has very little I

The 2" tubing may seem a bit extreme, but remember, our fitness function doesn't try to factor in material cost, it only tries to minimize the size of the conductor with respect to the other constraints. We could include a cost function ($) as a fifth constraint, and that should cause the GA to choose a smaller conductor.

The algorithm also generated a multiband loop that was simultaneously optimized for four different frequencies, 7.1, 10.125, 14.1, and 21.1 MHz. The loop diameter recommended by the GA was

If we want to bias the evaluation towards efficiency on 40m, we only need to adjust our fitness function to reflect that. I modified the fitness function for a single run, so that it considered conductor size for the lowest band only, and then evaluated all four of the fitness terms for all of the bands. This resulted in a modified version of

This demonstrates the power of adjusting the fitness function to emphasize certain design constraints over others. The optimization can still consider all of the target values, provided by the user, but the influence of a given term can be changed with respect to the others to give priority to one or more terms. In this latter example, the conductor size for the lowest band was given its normal weight, but for the higher bands, the weight given to conductor size was zero.

What this means for the antenna builder is that instead of concentrating on generating the specific measurements themselves, the designer can concentrate on the overall needs of the device, and the constraints within which he/she is trying to work. The machine does the work of finding the best solution within those constraints, allowing the human to focus on the bigger picture of how the antenna will actually be used.

These first experiments demonstrate that even a basic GA configuration can be used to generate very reasonable loop designs. Further, the algorithm can be made to do simultaneous optimization of a loop antenna for more than one band, if needed. This makes it a powerful tool for the loop experimenter to quickly build loops that should perform well, without having to hand-optimize the design parameters. The process can be expanded to very complex requirements sets, and the GA will crunch the numbers to determine the best solution.

During the design process, several parameters have been identified that could be used as inputs to the algorithm, being provided by the user as targets or constraints:

- Frequency of operation, or multiples thereof
- The desired maximum power level (W)
- The loop construction material (copper, aluminum, etc.)

- Efficiency
- Electrically small size
- Capacitor working voltage
- Maximum circulating current
- Bandwidth and Q
- Loop overall size/diameter
- Loop conductor gauge or diameter

The example results obtained so far certainly encourage further experimentation with this technology as a design tool.

The software used for this project is available for download. Source code is available as GZipped TAR files, and the compiled programs are meant to run at the command line on Linux or compatible system.

Source Downloads (Click Here)The software and source is being released under the GPL version 3. The license is available on the download page. |

LinksSmall Transmitting Loops - Details of the loops I have built for the contest bands. AutoCap - Automatic tuning control software for small loop antennas. |