Dy­namic Ra­dio Reg­u­la­tion with A.I.-As­sisted Spec­trum Eti­quette

A ra­dio spec­trum eti­quette pro­vides lo­ca­tion-, de­vice-, and sce­nario-de­pen­dent guide­lines to al­go­rithms for man­ag­ing and uti­liz­ing the ra­dio fre­quency spec­trum. De­vices fol­low such guide­lines by, for ex­am­ple, dy­nam­i­cally adapt­ing the wire­less com­mu­ni­ca­tion pa­ram­e­ters, any sig­nal char­ac­ter­is­tics, or the di­rec­tions and pat­terns of the used an­ten­nas. A spec­trum eti­quette fa­cil­i­tates a fair ac­cess to the ra­dio spec­trum, which helps avoid­ing in­ter­fer­ence, op­ti­miz­ing the ef­fi­ciency of ra­dio reg­u­la­tion, and fur­ther en­sur­ing a re­li­able com­mu­ni­ca­tion. Sev­eral al­go­rithms and tech­niques can be ap­plied to achieve these goals.

What is the Challenge

Na­tional reg­u­la­tory bod­ies, for ex­am­ple the Fed­eral Com­mu­ni­ca­tions Com­mis­sion (FCC), co­or­di­nate com­mer­cial ra­dio spec­trum us­age and the reg­u­la­tion of ra­dio emis­sions. The ra­dio spec­trum is di­vided into fre­quency bands, and li­censes for the use of fre­quency bands are given to -among oth­ers- tele­com op­er­a­tors. With li­censed fre­quency bands, op­er­a­tors are pro­vided with a long-term and ex­clu­sive right to use the ra­dio re­sources of the as­signed bands. Not al­ways will a band be fully used, and with an in­creas­ing num­ber of li­censes and in­creas­ing num­ber of ded­i­cated fre­quency bands, this leads to a high per­cent­age of ra­dio spec­trum be­ing scarcely used - or not at all.

A com­mon al­ter­na­tive to li­cens­ing is to pro­vide non-ex­clu­sive ac­cess per­mis­sions to cer­tain bands with­out such a li­cense. Bands made avail­able like this are known as li­cense-ex­empt or un­li­censed fre­quency bands. Be­cause of their gen­eral avail­abil­ity, un­li­censed fre­quency bands are used by many com­mer­cial and short-range wire­less com­mu­ni­ca­tion sys­tems. Wi-Fi, Blue­tooth, and ZigBee are pop­u­lar ex­am­ples.

How­ever, such ra­dio sys­tems are of­ten not de­signed to co­or­di­nate spec­trum with dis­sim­i­lar sys­tems, and as re­sult, they might in­ter­fere harm­fully with each other when used in close prox­im­ity.

Spectrum Etiquette

In an ideal sce­nario, spec­trum eti­quette rules are fol­lowed by all ra­dio sys­tems that op­er­ate in an un­li­censed band and share the spec­trum. The rules help reach­ing fair ac­cess to the ra­dio re­sources and a more ef­fi­cient use of the ra­dio spec­trum.

Select­ing the right pa­ram­e­ters can be lo­ca­tion-, de­vice-, and sce­nario-de­pen­dent. Al­go­rithms make de­ci­sions based on the ob­ser­va­tion of var­i­ous fac­tors in their en­vi­ron­ment. A de­ci­sion mak­ing al­go­rithm takes spe­cific char­ac­ter­is­tics of the in­volved ra­dio sys­tems into ac­count, in­clud­ing cov­er­age ranges, the num­ber of ac­tive de­vices, re­quired data rates, mo­bil­ity pat­terns. De­fault shar­ing rules are de­fined by reg­u­la­tion. Ad­di­tional fac­tors to con­sider are re­lated to the be­hav­ior of other sys­tems and the avail­abil­ity of cer­tain fre­quen­cies. The al­go­rithm uses all this in­for­ma­tion to ad­just its op­er­a­tional pa­ram­e­ters in real-time, such as fre­quency chan­nel, trans­mis­sion power level, and mod­u­la­tion and cod­ing scheme.

The goal of a spec­trum eti­quette is al­ways to max­i­mize the ef­fi­ciency of the ra­dio spec­trum us­age, while en­sur­ing fair ac­cess and min­i­miz­ing mu­tu­ally harm­ful in­ter­fer­ence.

Any shar­ing rule to ad­dress how un­li­censed de­vices should be­have should pro­mote greater co­ex­is­tence and spec­trum ef­fi­ciency in new un­li­censed bands. The ap­pli­ca­tion of such rules must al­low spec­trum in new un­li­censed bands to be used more in­ten­sively with a higher qual­ity of ser­vice than is pos­si­ble in the cur­rent un­li­censed bands, with­out lim­it­ing in­no­va­tion. The con­straints to be kept in mind when de­vel­op­ing shar­ing rules are:

• A shar­ing rule must not favour a spe­cific ra­dio stan­dard, limit the in­no­va­tion po­ten­tial of new gen­er­a­tions of tech­nol­ogy, or give an ad­van­tage to cer­tain ser­vices

• Min­i­mal­ist shar­ing rules should have demon­stra­ble ben­e­fits

• Shar­ing rules should not in­crease costs or com­plex­ity

Example Sharing Rule

The fol­low­ing ex­am­ple as­sumes a lis­ten-be­fore-talk medium ac­cess and a dif­fer­ent chan­nel band­width for the two dif­fer­ent sys­tems in­volved. The ex­am­ple rule is called “Fre­quency Chan­nel Clus­ter­ing “, and de­scribed as fol­lows: “When se­lect­ing a fre­quency chan­nel for op­er­a­tion, a ra­dio de­vice should se­lect a fre­quency chan­nel next to chan­nels that are al­ready used by sim­i­lar ra­dio sys­tems.”

The ob­jec­tive of this shar­ing rule is to max­i­mize the avail­abil­ity of spec­trum op­por­tu­ni­ties for other, com­pet­ing ra­dio de­vices. This rule is il­lus­trated in the fig­ure be­low. The ben­e­fit of the rule is clearly vis­i­ble. It will be help­ful to the broad­band de­vice(s) if nar­row­band de­vices would se­lect fre­quency chan­nels next to each other, as in­di­cated in the fig­ure.

Two dif­fer­ent ap­proaches to se­lect­ing a nar­row­band chan­nel for op­er­a­tion. Left: Fre­quency chan­nel se­lec­tion with­out eti­quette, ran­dom se­lec­tion. Here, the other ra­dio sys­tem op­er­at­ing with a broader chan­nel band­width will not find a free and un­used chan­nel. Right: With the pro­posed Fre­quency Chan­nel Clus­ter­ing eti­quette rule. In this case, two free broad­band chan­nels are avail­able for the other sys­tem.

Use of Machine Learning

De­vices that op­er­ate ac­cord­ing to dif­fer­ent ra­dio stan­dards with non-sim­i­lar mod­u­la­tion- and cod­ing schemes or pro­to­cols, are not di­rectly able to ne­go­ti­ate or co­or­di­nate the ac­cess to the ra­dio spec­trum. Pre­dic­tive an­a­lyt­ics and pat­tern recog­ni­tion are there­fore of­ten con­sid­ered to be used for an­tic­i­pat­ing fu­ture spec­trum al­lo­ca­tions from other de­vices. This would give a de­vice the means to re­act to the an­tic­i­pated al­lo­ca­tions, and se­lect re­sources ac­cord­ingly. The three main ap­proaches to learn from past spec­trum al­lo­ca­tions are re­in­force­ment learn­ing, su­per­vised learn­ing, and un­su­per­vised learn­ing. They dif­fer based on how data is pre­sented to the de­vice and how the de­vice learns to make pre­dic­tions or de­ci­sions.

In su­per­vised learn­ing, a de­ci­sion-mak­ing al­go­rithm is trained on a la­beled dataset, which means that each train­ing ex­am­ple is paired with an out­put la­bel. The goal is to learn a map­ping from in­puts to out­puts, al­low­ing the model to make pre­dic­tions on new, un­seen data based on this learned map­ping.

Un­like su­per­vised learn­ing, un­su­per­vised learn­ing in­volves train­ing the al­go­rithm on a dataset with­out any la­bels. The goal here is to dis­cover un­der­ly­ing pat­terns, struc­tures, or dis­tri­bu­tions in the data with­out any prior knowl­edge of out­comes.

Re­in­force­ment Learn­ing is fun­da­men­tally dif­fer­ent from both su­per­vised and un­su­per­vised learn­ing. In re­in­force­ment learn­ing, an agent learns to make de­ci­sions by per­form­ing ac­tions in an en­vi­ron­ment to achieve some goal. The agent re­ceives feed­back through re­wards or pun­ish­ments and learns over time to max­i­mize these re­wards.

A.I. Assistance

In a new ap­proach with the help of Ar­ti­fi­cial In­tel­li­gence (A.I.), rules can be rea­soned about, they might evolve over time, and they can be pro­posed and tested by the in­ter­act­ing ra­dio de­vices. Rea­son­ing en­gines as­sisted by Ar­ti­fi­cial In­tel­li­gence can use a sim­ple logic and might fur­ther im­prove the ef­fi­ciency of the orig­i­nal eti­quette rules pub­lished by the reg­u­la­tor. Iter­a­tion by it­er­a­tion, the spec­trum will then hope­fully be used more ef­fi­ciently.

Models de­rived from the the­ory of non-co­op­er­a­tive multi-stage games pro­vide an un­der­ly­ing frame­work for the rea­son­ing en­gines. They em­u­late the in­ter­ac­tion and com­pe­ti­tion for spec­trum, while each ra­tio­nal player (each in­ter­act­ing de­vice) aims to max­i­mize their own util­ity.

Back to: About Love­field – Overview