High Speed Trading Using Shortwave Radio

This is a paper I wrote for the final in a class on communication networks while at RIT. I go through challenges of using shortwave radio and propose a solution for a low latency, highly reliable data link over shortwave radio. (Paper was submitted on December 7, 2020)

Abstract—High Speed Trading companies are always trying to stay on the cutting edge of computing and networking to enable them to make more trades faster. One method of high speed trading, arbitrage, involves buying and selling stocks at different locations. This requires the two sites to be connected by a very low latency link. Current methods involve using microwave links but that does not work for links across the Ocean. Using shortwave radio is a proposed method to make this link. This paper looks at issues with shortwave radio like propagation and modulation and proposes solutions. Using multiple frequencies and an encoding mode with lots of error correcting will allow a very low latency, low throughput system to operate with high reliability over the shortwave band at all times.

I. INTRODUCTION

High speed trading is a method of trading where companies buy and sell stocks very quickly using algorithms. One trade can happen in as little as 64us [1]. This means the latency between the servers running the algorithms and the stock exchange servers is very important. Large data centers are built near the stock exchanges so there is very low latency. One method high speed traders use is buying stocks at one exchange, and then selling them almost instantly at another exchange before the prices change [2]. This is called Arbitrage. The company could be selling the stocks for only a few cents more than they were bought, but when this is done thousands of times, large gains can be made. For this to work the company has to have fast connections to at least two stock exchanges. A common two to link are the New York Stock Exchange and the Chicago Stock Exchange. Companies have built microwave relays between the two stock exchanges to have the fastest possible link [3]. These links have reduced the latency from 7.95 ms to 4.2 ms. This works as when the signal is traveling as a radio wave, it is traveling at the speed of light, as opposed to only 68% of the speed of light through fiber optics. These links work very well, but one limitation is they are limited to only 70 km jumps due to the curvature of the Earth. If a company wanted to link the Chicago Stock Exchange and the London Stock Exchange, a different method would need to be used.

Currently North America and Europe are linked by undersea cables. One cable, the AEC-1, connects NYC to London [4]. It takes 67.86 ms for a signal to travel from one end to the other. If the signal could travel the speed of light, it would only take 18.63 ms. A proposed solution is to use shortwave radio to link the two continents.

II. CURRENT HIGH SPEED TRADING WITH SHORTWAVE

There are no high speed trading companies that are publicly disclosing that they are experimenting using shortwave radio. This is because there is a lot of money involved in this and companies do not want others to steal their work. By using FCC license databases as well as satellite imagery, it is possible to get an idea of what companies are trying. One such license, WH2XVO, is linked to Virtu Financial, a provider of financial services [5]. They have a license for a variety of frequencies from 6 to 24 MHz with up to 50 kW of power. The largest bandwidth they have is only 1 MHz. Information about transmitter and antenna equipment is redacted. Many of the licenses are for antennas setup in Illinois near Chicago or in New Jersey near NYC. This makes sense as the sites would be near either the Chicago Stock Exchange or the New York Stock Exchange. One person found an old cell phone tower outside Chicago with some shortwave antennas and a microwave link. He figured out the site was being used for high speed trading with the shortwave antennas pointing towards Europe [6].

III. SHORTWAVE

Shortwave radio is the band from 3 to 30 MHz. It has wavelengths from 10 to 100 m [7]. These signals will bounce off the ionosphere allowing them to reach very far distances. This does come with some challenges though. The first challenge is propagation. How the signals bounce off the ionosphere changes based on the time of day as well as the solar cycle. The solar cycle is an 11 year cycle the sun goes through. Over the cycle the amount of sun activity changes. When the sun is more active, shortwave propagates better [8]. Different frequencies propagate better at different times of the day as well. The second challenge is there is limited bandwidth, so special data modes must be used. A microwave link can use 56 MHz of bandwidth, while many shortwave digital modes only use a maximum of 2.4 KHz.

IV. PROPAGATION

The first issue that needs to be solved is propagation. Using Voice of America Coverage Analysis Program (VOACAP) propagation over different times of the day and frequencies can be looked at [9]. An eight element Yagi antenna is set up in Chicago and London mounted at 198 feet. 20kW of power is used. The easiest parameter to look at is the Reliability Factor. This is the percent of the time that a given hour will have a reliable signal. This is based on the required Signal to Noise Ratio (SNR) for a given mode. Required SNR is based on both the mode as well as the bandwidth being filtered. For a Single Side Band (SSB) Signal a SNR of 6 dB is required. With a digital mode, a SNR as low as -32 dB is needed. Both numbers are based on a 2500 Hz bandwidth [10]. Looking at 7.1 MHz, the propagation is very good for most of the day, except around 1500 UTC (10am EST). Figure 1 shows the reliability of AM at 7.1 MHz. It has a minimum of 55% reliability at 1500 UTC. This chart was predicted on December 6, 2020. The reliability can change depending on the time of year. The same chart was generated on November 18, 2020 and is shown in Figure 2. It has a minimum reliability of 0% at 1500 UTC. Here it can be seen how quickly propagation can change. This can cause issues for a high speed trading company, as they need very high up time. One solution for this is to use multiple frequencies. By using a frequency that is better during the day, like 7.1 MHz, and one that is better at night, like 10.1 MHz, allows at least one to work well at all times. A disadvantage is two different antennas are needed. Figure 3 shows the reliability of AM at 10.1 MHz predicted on December 6, 2020. It has a maximum of 98% at 1400 UTC and a minimum of 0% at 0600 UTC. By using 40 m and 30 m (7.1 MHz and 10.1 MHz) high reliability is available for the entire day. This is true for December 6 2020, but could change with the solar cycle. To ensure good reliability even when propagation is not good, a digital encoding mode that works with a low SNR should also be used.

V. DIGITAL MODES

The second issue is the limited bandwidth available in the shortwave band. The limited bandwidth means throughput will be limited. Luckily this is not a huge issue as just the stock symbol, buy or sell, and the amount to buy or sell are needed to be transmitted. A stock symbol is at most 5 characters [11]. If this is converted to ASCII it is 8 bits per letter, so 40 bits. Buy or sell can be a single bit, and then the amount of stocks to buy or sell. With 14 bits this gives a max transaction of 16,383. This is a total of 55 bits per transaction. The next problem is how to encode these 55 bits.

In Amateur Radio there are a variety of digital modes meant for low power use on the HF bands, which includes shortwave [10]. These are a good starting point for developing a method for high speed trading. The first mode is WSPR (pronounced “whisper”). This is a beacon mode designed for testing propagation. It has a very low bandwidth of only 6Hz and can be received with as low of a SNR as -32 dB. The tradeoff is that it is very slow, sending only 50 bits in 110.6 seconds, giving a bit rate of 0.45 bps. The mode has lots of error correcting which is one reason the bit rate is so slow. Another common mode is FT8. This has a bandwidth of 50 Hz and can be received down to -21 dB. It can send 74 bits of user data over 12.6 seconds, giving a bit rate of 5.87 bps. Another version called FT4 is faster, but needs a slightly high SNR of -17.5 dB. It transmits 74 bits over 5.04 seconds, giving a bit rate of 14.68 bps. The table below shows WSPR, FT8, and FT4 as well as some other common digital modes in amateur radio.

From this table, it can be noted that the higher the bandwidth, the faster the bit rate is. This makes sense and is the main limitation with shortwave frequencies. Out of these options, JT9 seems very promising. There are different versions of the protocol for different bandwidths and all can be decoded with a SNR around -20 dB.

Looking back at the WH2XVO license, they have access to 235 KHz of bandwidth from 6.765 to 7 MHz as well as 1 MHz of bandwidth from 10.15 to 11.175 MHz. At both frequencies they can transmit at 29.5 kW. This gives them plenty of room to either up the bandwidth on the transmitted signal to get a faster bit rate, or to use multiple transmitters in parallel.

From looking at the table, it can be seen that for the same mode, if the bandwidth is doubled, then the bit rate also doubles. If JT9 is used and expanded to use the full 235 KHz of bandwidth available at 7 MHz, then it will have a throughput of 130 times JT9H, or 22117.4 bps. This is the same as 22.12Kbps or 0.02 Mbps. This is still very slow, but approaching the point where standard TCP/IP could be used. Doing TCP/IP over this link would add more headroom and slow down the already slow link. Using a custom interface over the shortwave link and converting to TCP/IP at either end will be more efficient than trying to use TCP/IP over shortwave.

These amateur radio modes include lots of error correcting, but do not include any provisions for acknowledgements or retransmitting. This is something that would have to be added and would again reduce the effective bit rate. One possible solution is to use two sets of parallel transmitters and receivers. This would allow for full duplex communication as well as a second channel for acknowledgements.

VI. LOCATION

Currently shortwave stations in the US for high speed trading are mostly outside Chicago [6]. This is because there is a lot of land for relatively cheap and it is close to a stock exchange. This does mean that the radio signal has to travel further to reach England than if it was closer to the coast. In England the stations are often located on the South West tip. There is a microwave link that then connects to London. One option to improve reception would be to locate the shortwave station in Nova Scotia. This would result in being 2,250 km closer to England. This would involve building a microwave network from Nova Scotia to the US and linking it in to the existing network.

Figure 4 shows a map with pins at Chicago, NYC, Nova Scotia, and England. From the map it is easy to see how making the station in Nova Scotia would greatly reduce the distance the radio signal would have to travel.

VII. OVERALL PLAN

The overall plan for a system that uses shortwave radio to improve latency for high speed stock trading would involve a system using four different antennas. Two would be on the 40 M band and the other two on the 30 M band. By using both bands the best reliability would be available over the entire day. Two transceivers would be used, with a switch to control which antenna is being used. This system would be duplicated on the other side as well. One transceiver would be transmitting the buy and sell information, while the other is receiving acknowledgements from the other side. On the other side one would be receiving the buy and sell orders and the other would be transmitting the acknowledgments. This allows for full duplex communication. Both would be using a modified version of JT9 to use all available bandwidth. At each side the transmitted and received buy and sell orders would be translated into TCP/IP to be sent along existing network infrastructure to the stock exchange. The system would only send acknowledgments. If an acknowledgment is not received within a set time frame, then the message is sent again. One limitation of this setup is that there is no encryption other than obscurity. If someone, like a competitor, were to figure out how to demodulate the signals, then they could make their own trades off the information and possibly take away profit. Adding encryption would reduce throughput, but might make sense to protect the data.

VIII. CONCLUSION

From this paper it can be seen that a shortwave link that is both low latency and highly reliable is possible. Although no company has publicly stated that they are using shortwave for high speed trading, it is only a matter of time until one does. With many companies already experimenting with shortwave radio links, it is possible that shortwave radio will be making a comeback for low latency network applications.

REFERENCES

[1] Eric Reed, What is High-Frequency Trading?, SmartAssest, 01/14/2020, [Online]. Available: https://smartasset.com/investing/high-frequency-trading

[2] James Chen, Arbitrage, Investopedia, 02/01/2020, [Online]. Available: https://www.investopedia.com/terms/a/arbitrage.asp

[3] Gregory Laughlin, Anthony Aguirre, Joseph Grundfest, Information Transmission Between Financial Markets in Chicago and New York, Financial Review, 02/26/2013

[4] AEC-1/AEConnect, Submarine Cable Networks, [Online]. Available: https://www.submarinenetworks.com/systems/trans-atlantic/aeconnect

[5] Experiments Look to Leverage Low-Latency HF to Shave Microseconds off Trade Times, ARRL News, 06/20/2018 http://www.arrl.org/news/experiments-look-to-leverage-low-latency-hf-to-shave-microseconds-off-trade-times

[6] Bob Van Valzah, Shortwave Trading | Part I | The West Side Chicago Tower Mystery, Sniper in Mahwah & Friends, 05/07/2018, [Online]. Available: https://sniperinmahwah.wordpress.com/2018/05/07/shortwave-trading-part-i-the-west-chicago-tower-mystery/

[7] Short Wave Frequency Bands, shortwave-wave.info, [Online]. Available: https://short-wave.info/index.php?feature=frequencies

[8] What is the Solar Cyle?, NASA Science, 09/10/2020, [Online]. Available: https://spaceplace.nasa.gov/solar-cycles/en/

[9] Jari Perkiomaki, VOACAP Overview, VOACAP, [Online]. Available: https://www.voacap.com/overview.html

[10] Joesph H. Taylor, WSJT-X User Guide, Princeton, [Online]. Available: https://physics.princeton.edu/pulsar/k1jt/wsjtx-doc/wsjtx-main-2.2.2.html

[11] Adam Hayes, Stock Symbol (Ticker), Investopedia, 09/29/2020, [Online]. Available: https://www.investopedia.com/terms/s/stocksymbol.asp