C                                                                                                                         ...
C                                                                                                                         ...
C                                                                                                                         ...
C                                                                                                                         ...
C                                                                                                                         ...
C                                                                                                                         ...
C                                                                                                                         ...
C                                                                                                                         ...
C                                                                                                                         ...
Auction based resource allocation in cognitve radio (get more insights from: http://trends-in-telecoms.blogspot.com/)
Auction based resource allocation in cognitve radio (get more insights from: http://trends-in-telecoms.blogspot.com/)
Auction based resource allocation in cognitve radio (get more insights from: http://trends-in-telecoms.blogspot.com/)
Auction based resource allocation in cognitve radio (get more insights from: http://trends-in-telecoms.blogspot.com/)
Upcoming SlideShare
Loading in …5
×

Auction based resource allocation in cognitve radio (get more insights from: http://trends-in-telecoms.blogspot.com/)

771 views

Published on

auction based resource allocation in cognitve radio

Published in: Technology, Business
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
771
On SlideShare
0
From Embeds
0
Number of Embeds
24
Actions
Shares
0
Downloads
0
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide

Auction based resource allocation in cognitve radio (get more insights from: http://trends-in-telecoms.blogspot.com/)

  1. 1. C qM IEEE M ommunications q qM Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page MqM q Qmags THE WORLD’S NEWSSTAND® COMMUNICATIONS NETWORK ECONOMICS Auction-Based Resource Allocation in Cognitive Radio Systems Yang Zhang, Dusit Niyato, Ping Wang, and Ekram Hossain ABSTRACT works (e.g., network operators) for long-term use. The primary networks contain primary base Auction theory, as a subfield of economics, stations and primary end users. The primary base provides useful tools to model, analyze, and stations own the licensed spectrum and serve optimize radio resource management in cogni- their primary end users (e.g., mobile phones) tive radio environments. By using an auction, within coverage. The secondary base stations radio resources such as subchannel, time slot, dynamically request and access available spec- and transmit power can be allocated among trum (i.e., spectrum hole or spectrum opportuni- licensed and unlicensed users in the system, fol- ty) from the primary base stations, and use the lowing market laws. Due to the flexibility of spectrum to serve their secondary end users. For mechanism design, there are various auction the rest of this article, we use primary user to mechanisms that have been applied to cognitive represent both primary base stations and prima- radio systems with different characteristics. In ry end users, and secondary user to represent this article, we first provide an overview of the both secondary base stations and secondary end basics of general auctions. Then the motivations users if there is no ambiguity. and specific design issues in applying auctions to Secondary users make payments to primary wireless network architectures and protocols are users as an incentive for the primary users to discussed. Then we review the state of the art in lease the licensed spectrum, hence making cog- the use of auction theory and mechanism design nitive radio systems analogous to real markets. in cognitive radio networks. This will enable the The primary and secondary users can trade readers to have a general view of auction funda- resources (e.g., spectrum) under certain market mentals, as well as the recent development and regulations. Economics and business methods applications of auction theory in the emerging are therefore natural to design and analyze radio cognitive wireless networks. resource allocation and spectrum management in a cognitive radio system. INTRODUCTION The auction theory [3, 4] developed in eco- nomics has recently been applied to solve vari- Wireless and mobile technologies are growing ous problems in engineering. An auction process rapidly, leading to much more complex spectrum works as follows. First, buyers submit bids for usage, where the supply of and demand for radio purchasing commodities, and sellers submit asks spectrum vary dynamically. The conventional to sell commodities. An auction commodity fixed spectrum allocation approach has been being traded may be an actual item or a service. shown to be unsuitable for flexible and efficient Each bid/ask contains information that indicates use of the scarce radio spectrum. Cognitive radio the buyer’s/seller’s preferences, requirements, systems [1, 2] are therefore designed to provide and requests for the commodities to be traded. flexibility and efficiency in spectrum usage by Generally, a bid needs to at least express a bid allowing unlicensed users (i.e., secondary users) price and quantity of commodity to be pur- to opportunistically access the spectrum allocat- chased. Similarly, an ask should contain an ask- ed to licensed users (i.e., primary users). ing price and quantity of commodity to be sold. A cognitive radio system can be either single- The price included in a bid/ask might not be the hop or multihop. For simplicity, we discuss a sin- buyer/seller’s real valuation, which is a private gle-hop structure; that is, there is only one hop value assessed by the buyer/seller on the com- between the source and the destination. In the modity to be bought/sold. There is also an inter- theoretical studies and industry applications, a mediate agent referred to as an auctioneer who hierarchical model could be used to describe a conducts the auction and clears the market by cognitive radio system, as shown in Fig. 1. Note matching bids and asks. Thereafter, every that the layered model is just one possible matched buyer-seller pair must deal at a clearing method to be applied. Three main components price (or dealing price). In some auctions, there is in the cognitive radio model are spectrum author- only one seller, and this seller can perform a ities, primary networks, and secondary networks. role of an auctioneer. As a result, in such a case, Spectrum authorities are the ultimate spectrum the terms auctioneer and seller can be used rights owners, such as governments, who official- interchangeably. ly lease out spectrum licenses to primary net- In the context of the aforementioned layered 108 0163-6804/12/$25.00 © 2012 IEEE IEEE Communications Magazine • November 2012C qM IEEE M ommunications q qM Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page MqM q Qmags THE WORLD’S NEWSSTAND®
  2. 2. C qM IEEE M ommunications q qM Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page MqM q Qmags THE WORLD’S NEWSSTAND® cognitive radio model, the buyers1 are secondary users who can use the spectrum by paying for the spectrum, and the sellers (e.g., primary users) provide the radio resources to buyers and receive monetary gains from buyers by selling Spectrum authority unused or underutilized radio resources. The commodities are the radio resources (e.g., band- width, time slots, and rights to transmit data in the network). In addition, the auctioneers can charge a commission fee from arranging auc- Primary ... Primary Primary tions for primary and secondary users to gain base station base station user revenue. Through the auction, primary users, secondary users and auctioneers have motiva- tions to participate to gain benefits in a cognitive ... radio system. Primary This article presents a survey on the applica- user tions of auction theory in cognitive radio sys- Secondary Primary network base station tems. First, we provide an overview of fundamental concepts, categorization, and objec- Secondary user tives of auctions. Next, we discuss the issues related to modeling cognitive radio systems as an auction market. Different auction mechanisms Primary network Secondary network used in the literature to model cognitive radio systems are reviewed. Several research directions are then outlined before the article is concluded. Figure 1. Cognitive radio system architecture. FUNDAMENTALS OF the seller gains positive profit. The final dealing price will at least fall in between a seller’s valua- AUCTION THEORY tion and a buyer’s valuation to close the trade. Auction mechanisms can be categorized based The players are assumed to be rational and tend on different criteria. An auction could be open- to strategically optimize their own profits. cry or sealed-bid. In an open-cry auction, the An ideal objective of an auction mechanism buyers and sellers, respectively, publicly reveal as a game is to analyze dominant strategy equi- information about their bids and asks during the libria of the buyers and sellers. A dominant auction process. On the other hand, in a sealed- strategy will provide the best payoff in any case, bid auction, the information is private, and only so that a rational player will stick to the strategy revealed to the auctioneer. An auction could be regardless of other players’ strategies. The domi- single-sided or double-sided. In a single-sided nant strategy design is suitable for sealed-bid auction, either buyers or sellers submit their bids auctions, where each buyer/seller knows nothing or asks, respectively. In contrast, in a double- about bidding strategies of other buyers/sellers. sided auction, both buyers and sellers submit For example, the widely studied truthful auction their bids and asks. An auction could be forward mechanisms are to encourage the buyers to bid or reverse. In a forward auction, the buyers bid exactly their true valuations as their dominant for commodities from seller(s). Sellers ask to sell strategies in the auction. commodities to buyer(s) in a reverse auction. An The Nash equilibrium is also a common solu- auction could be static or dynamic. In a static tion concept in auction games, ensuring that auction mechanism (e.g., a sealed-bid auction), none of the players will unilaterally change its participants generally do not update their strate- strategy given that other players keep their gies based on external information over time. In strategies fixed. It is important to check the exis- a dynamic mechanism, the buyers and sellers tence and/or uniqueness of the Nash equilibrium have a chance to update their strategies by col- so that the auction will finally terminate and lecting information from other participants and generate desirable results. Achieving a Nash the revealed past results. An auction could be equilibrium might require the players to have single-unit or multi-unit where only one or mul- knowledge about other players in the mecha- tiple commodities are involved, respectively. The nism, which could be impossible in many sealed- major types of auction are shown in Fig. 2, and bid auctions. the details are given in Table 1. Refer to [4] for a detailed tutorial on auction theory and models. AUCTION AND OTHER MARKET APPROACHES: A COMPARISON AN AUCTION AS A GAME Game theory is a typical mathematical tool used As one of the trading and pricing schemes [5] to analyze behaviors of buyers and sellers (i.e., that can be applied to resource management, an players) in an auction, for example, to submit auction is often compared with other market the optimal bids and asks. In the bid from the approaches, such as dealer market and posted buyer, the bid price cannot exceed the valuation price market. of the commodity for which the buyer bid, so the In a dealer market, there is a market maker buyer has positive gain after auction. Similarly, (or an exchange) who deals with both buyers and 1In the rest of article, we the ask price from the seller must be higher than sellers of commodities, and accepts their bids may use “buyer” and the valuation of the spectrum to be sold; thus, and asks in any case if the prices are acceptable. “bidder” interchangeably. IEEE Communications Magazine • November 2012 109C qM IEEE M ommunications q qM Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page MqM q Qmags THE WORLD’S NEWSSTAND®
  3. 3. C qM IEEE M ommunications q qM Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page MqM q Qmags THE WORLD’S NEWSSTAND® A posted price market is similar to Seller Seller Seller an auction. There is Seller Seller no official exchange Seller Seller Seller Seller as the market maker, and the trading will only be done if a bid Auctioneer Auctioneer Auctioneer and an ask are matched. However, the bid/ask prices Buyer Buyer Buyer Buyer Buyer Buyer Buyer submitted are public Buyer Buyer (a) (b) (c) and can be seen by all the market participants. ... ... (d) Figure 2. Different auction types, where arrows indicate commodities and money transfers among entities in an auction: a) a forward auction with a single seller; b) a reverse auction with a single buyer; c) a double auction; d) a double auction in a complex market with multiple trading agents. Therefore, the market maker is the only coun- the average value of all reference prices as its terparty of both buyers and sellers in a dealer own bid prices. Such a market is similar to an market, which means the market maker can open-cry auction. However, on the contrary, accept a bid from a buyer even if there is no there may not be such significant market partici- matched seller in the market at the moment pants in an auction. Moreover, sealed-bid auc- (i.e., counterparty risk). However, the dealer tions do not support publicly posting prices. market is based on the assumption that there are enough active buyers and sellers in the market to provide trades of high liquidity. Otherwise, the ISSUES IN DESIGNING AUCTION- market maker might fail to allocate commodi- BASED APPROACHES FOR COGNITIVE ties/money to buyers/sellers immediately because there are no sellers/buyers. In contrast, the allo- RADIO SYSTEMS cation can be executed right after the moment a RADIO RESOURCES FOR AUCTION bid and an ask are matched in an auction. Also, it is a critical issue for the market maker to Spectrum allocation in cognitive radio systems is determine the price to accept the incoming bids more often referred to as dynamic spectrum and asks, respectively. Failing to do so, the mar- access (DSA) or opportunistic spectrum access ket maker may be exposed to the chance of the (OSA). In DSA, first the spectrum holes must be loss of money. identified (i.e., spectrum exploration), and then A posted price market is similar to an auc- the secondary users can access these spectrum tion. There is no official exchange as the market holes (i.e., spectrum exploitation). An auction maker, and the trading will only be done if a bid can be applied between spectrum exploration and an ask are matched. However, the bid/ask and spectrum exploitation of DSA. The different prices submitted are public and can be seen by types of radio resources for auction are as fol- all the market participants. Normal participants lows. in a posted market might trust and take the Subchannel: The frequency bands unused by quotes proposed by some “significant” partici- the primary networks can be divided into sub- pants (e.g., large financial institutions) as the channels. The secondary users can bid for sub- reference prices. For example, a buyer can use channels for their data transmission without 110 IEEE Communications Magazine • November 2012C qM IEEE M ommunications q qM Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page MqM q Qmags THE WORLD’S NEWSSTAND®
  4. 4. C qM IEEE M ommunications q qM Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page MqM q Qmags THE WORLD’S NEWSSTAND® Auction name Key features Example scenario Reference • Open-cry bidding process English auction A government auction for spectrum licenses [3] • Ascending price • Bid prices quoted by sellers Selling commodities that decay or expire Dutch auction [3, 29] • Descending price easily • Valuations are private to competitors Vickrey auction • Winning bidder pays the 2nd highest price A single-commodity sealed-bid auction [3, 9, 20] • Truthfulness guaranteed • Generalized Vickrey auction Vickrey-Clarke-Groves A multi-commodity sealed-bid spectrum [6, 20, 27, • Truthfulness guaranteed (VCG) auction license bidding, truthfulness required 30] • Often used in theoretical studies • Ask/bid direction of original auction is Reverse English auction where the seller Reverse auction [4, 32] reversed decreases asks to compete for buyers Real market with many sellers, bidders and • Multiple sellers and bidders Double auction agents as auctioneers, e.g., a stock [4, 27, 28] • Complex market exchange • Multiple auction commodities A buyer needs a consecutive bundle of fre- Combinatorial auction • Players only value bundles of commodities [12, 32, 33] quencies in a spectrum auction most Table 1. Summary of general auction types and their features. interfering with the transmission by primary resources may be traded in two layers of mar- users (Fig. 3a). The subchannels can be homoge- kets. In an auction held in a primary spectrum neous (i.e., perfectly substitutable) if all the sub- market (i.e., higher-layer market), sellers are channels have the same quality for buyers to spectrum authorities, and buyers are primary transmit data. Otherwise, the subchannels are users [7]. In a secondary spectrum market auc- heterogeneous if they have different qualities for tion (i.e., lower-layer market, e.g., [8]), primary different buyers (e.g., due to frequency selective users become sellers, and secondary users are fading). buyers. A secondary spectrum market is general- Time slot: Spectrum can also be divided in ly much more interactive and dynamic than a the time domain. An available time period to primary spectrum market. access a channel is divided into time slots as auc- tion commodities (Fig. 3b). Different time slots SPECIFIC DESIGN ISSUES may have different valuations in the view of buy- Currency and Payment Design — Currency, ers due to time-varying channel quality. generally referred to as money, is an incentive Transmit power or signal-to-interference- for the radio resource sellers (e.g., primary plus-noise ratio (SINR) level: The transmission users) to participate in an auction. In a cognitive by secondary users may cause interference to the radio system, through auction, the sellers sell the primary users. A primary user can set a maxi- radio resources to buyers (e.g., secondary users). mum transmit power level for secondary users. Once the radio resources are allocated to the The secondary users as buyers have to pay high- buyers, they make a payment to the sellers using er price for higher transmit power (due to higher currencies. In various spectrum auction designs, interference caused to the primary users). An currency can be in the form of real cash (e.g., as auction for transmit power for secondary users is in governments’ spectrum licenses sale [7]) or shown in Fig. 3c. fictitious currency [9, 10]. Network service: In a heterogeneous network We illustrate the designs of currency and pay- with multiple radio transmission technologies, as ment in auctions by adopting the aforemen- shown in Fig. 3d, different wireless access net- tioned layered cognitive radio model [1, 2] shown works (e.g., wideband code-division multiple in Fig. 1. A primary spectrum auction is usually access [WCDMA] and IEEE 802.11b) can be conducted by spectrum authorities. In such auc- seen as commodities for an auction [6]. This is tions, the currency can be real cash. When the referred to as joint radio resource management auction process becomes highly dynamic in a (JRRM), where each wireless service provides a secondary spectrum auction, it may not be prac- certain quality of service (QoS) and charges a tical to make every transaction using real cash. certain price. On one hand, network users as the Therefore, in the secondary spectrum auction, buyers can compete for the available wireless fictitious currency may be used [9], which is service through an auction. On the other hand, designed as a signal representative of some different wireless services can compete to attract amount of real currency. Compared to real cash, the users to gain revenue through the auction. the fictitious currency is easy to circulate in the In cognitive radio systems, the spectrum wireless environment. However, there is a chance IEEE Communications Magazine • November 2012 111C qM IEEE M ommunications q qM Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page MqM q Qmags THE WORLD’S NEWSSTAND®
  5. 5. C qM IEEE M ommunications q qM Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page MqM q Qmags THE WORLD’S NEWSSTAND® A simple language is Auction period Time slot preferable in a typi- Channel cal auction due to Seller Seller Subchannel small overhead. Auction control Auction control However, the simple language may not ... ... Buyer Buyer be capable enough Buyer Buyer ... ... to carry all necessary information, and an Buyer Auctioneer Auction control Buyer Auctioneer Auction control interpretation of auc- (a) (b) tion message using simple language can be inefficient due to ambiguity or lack of Power level 3 Buyer a clear structure. Seller Buyer Bluetooth Po w er Auctioneer lev el 2 Buyer Pow Buyer e Buyer r le Buyer vel 1 Buyer 802.11b WCDMA (c) (d) Figure 3. a) A spectrum auction for subchannels; b) a spectrum auction for time slots; c) a spectrum auc- tion for levels of transmit power; d) an area covered by different types of network service (e.g., the buyer in the middle is covered by Bluetooth, 802.11b, and WCDMA simultaneously). that such fictitious currency can be counterfeited sively but efficiently in indicating spectrum buy- by sellers or buyers. Generally, the payment ers’ needs. A simple language (e.g., with a plain transaction can be made through a centralized structure) is preferable in a typical auction due entity (e.g., base station). Alternatively, to to small communication overhead. However, the reduce the communication overhead with the simple language may not be capable enough to central entity, it can be performed by the sellers carry all necessary requirements and information and buyers themselves in a distributed manner. (e.g., in combinatorial auction), and an interpre- In addition to the payment using currency, a tation of auction message using simple language barter-like trading mechanism can be employed can be inefficient due to ambiguity or lack of a in an auction. Specifically, spectrum sellers and clear structure. buyers offer their redundant resources in During an auction period, each secondary exchange for demanded resources. In [10], the user can submit just one bid message, or multi- primary user can request the secondary users to ple bid messages over time. In each bid message relay the data transmission to the destination, submitted, there can be a single request included reducing the transmit power of the primary user (i.e., single-bid). For example, a bidding pair and improving the transmission quality. In ( F b i , p i ) indicates that the secondary user b i return, the secondary user obtains the rights to needs a range of frequency F bi at price pi. The access licensed radio resource of the primary SALSA model in [11] also introduces an exam- user. In this case, the transmit power of the sec- ple of single-bid language. Each bid message is a ondary users used for relay transmission of the five-tuple set containing a single resource request primary user is bartered with the radio resource for time slots to be used for data transmission. given by the primary user. Single-bid bidding language has relatively simple structure. However, each bid message can con- Bidding Language Design — Bidding lan- tain a complex structure of several single-bids, or guages are the expressions to represent the atomic bids. A set of resource amount-price information exchanged among buyers and sell- pairs and logical operators can be included in ers. The primary and secondary users must use the bid message. This is referred to as the multi- the same language to exchange auction mes- bid bidding language. For example, an “OR bid” sages. Typically, the bidding message contains [12] S can express the bidder’s need for any sub- secondary users’ bid prices as well as other per- channel f i at price p i , i = 1, …, M. This com- formance and resource requirements. pound bid S contains M atomic bids and an OR Bidding languages should be designed expres- strategy, that is, 112 IEEE Communications Magazine • November 2012C qM IEEE M ommunications q qM Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page MqM q Qmags THE WORLD’S NEWSSTAND®
  6. 6. C qM IEEE M ommunications q qM Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page MqM q Qmags THE WORLD’S NEWSSTAND® S = (f1, p1) OR (f2, p2) OR … OR (fM, pM). Emphasizing fairness, for example, [16] devel- ops a fairness index in spectrum resource auctions Commodities in con- Bidding language can be compressed into a that is a function of all secondary users’ utilities more compact form to support real-time spec- and the number of secondary users. A smaller ventional auctions trum trading. After the auctioneer collects value of fairness index indicates a more unfair are generally not enough bids, whether from a single-bid or multi- mechanism (i.e., spectrum allocation results are reusable. When a bid process, a demand curve can be constructed. more in favor of “rich” buyers). On the contrary, The demand curves can be expressed as analyti- when the value of the fairness index is large, the commodity is cal functions. Bidders just submit messages con- allocation scheme tends to distribute resources obtained by one taining these functions or the parameters of evenly among all buyers. The usage of a fairness predefined functions. In [13], a secondary user index is twofold. First, by maximizing or minimiz- winner, it cannot be approximates a radio resource demand curve as a ing the fairness index, the auction mechanisms used by other bid- piecewise linear price-demand (PLPD) function. can decide whether to achieve a fairness or an ders unless the win- The secondary users then send the parameters of efficiency objective. Second, the fairness index this PLPD function to the auctioneer to express can also be used to measure quantitatively ner releases it. various types of demand preference. In a more whether an existing auction is fair or not. As one However, in a wire- general case, the secondary users do not need to of the objectives, the auction mechanism pro- establish and maintain their own function as bid- posed in [16] maximizes the value of the fairness less system, especial- ding language. Instead, the auctioneer (e.g., pri- index. ly in a cognitive radio mary user) has a collection of predefined In [8], mentioned earlier, the fairness metric system, as a com- functions for secondary users. The secondary is not even explicitly defined. But the auction users choose one of these predefined functions to model still guarantees basic fairness. In the long modity, the radio represent their resource requests. Since the func- run of the auction process, none of the rational resource can be used tion can be represented in a much more compact secondary users will bid all the time due to the form than that of general auction languages, the cost of entrance fee. This improves the chance by multiple users communication overhead is significantly reduced. of other secondary users with lower valuations to simultaneously. access the channel, and thus guarantees the fair- Efficiency of Auction Mechanisms — The ness to some extent. efficiency objective [14] in auctions is generally referred to as allocative efficiency, also known as Spectrum Reuse and Collision — Commodi- Pareto efficiency, where the efficiency requires ties in conventional auctions are generally not the auction commodities to be allocated to those reusable. When a commodity is obtained by one who value them most. Therefore, it is a specific winner, it cannot be used by other bidders unless and measurable metric when designing an auc- the winner releases it. However, in a wireless tion for a cognitive radio system. system, especially in a cognitive radio system, as We take [8] as an example, where efficiency a commodity, the radio resource (e.g., subchan- is the main objective of the proposed auction nel) can be used by multiple users simultaneous- mechanism. The secondary users are charged ly. For example, a subchannel can be used by with an “entry fee” once they choose to enter different users at the same time if these users do the market to make bids. However, the sec- not interfere with each other. This subchannel ondary users are rational and can observe past can be considered a reusable auction commodi- auction results. Therefore, they autonomously ty, and there can be multiple winners of the decide whether to join the market before every same commodity. auction period. For example, the secondary The authors in [17] model an auction with users with relatively low valuation on radio spectrum reuse as a multi-winner auction. Sec- resources are less likely to bid, avoiding the cost ondary users who bid for the same subchannel of the entry fee. In such a spectrum market, the and do not interfere with each other are grouped efficiency objective can be achieved by market as a virtual bidder. Each virtual bidder acts as a competitions among rational self-interested sec- normal bidder, and its virtual valuation is calcu- ondary users. This is similar to the “invisible lated by summing all valuations on that subchan- hand” theory in economics [15]. However, nel. The secondary users in a virtual bidder can achieving efficiency by competition may rely on share the same subchannel without conflict. three preconditions. First, the quantity of active Therefore, they are all winners of the auction if participants (i.e., price takers) in the market their bid is the highest. Note that virtual bidders must be large enough so that the strongest par- are not fixed since the availability of the sub- ticipants can be gradually chosen out of competi- channel to different secondary users can dynami- tion. Second, the auction should be conducted cally change (e.g., due to mobility and channel for several rounds to converge to a stable mar- fading). In [18], the virtual bidder concept is ket. Third, the market participants should considered as a coalition of multiple winners. dynamically adjust their strategies based on the Thus, such an auction is essentially a cooperative information they are able to access. game [19]. By using the Shapley value [19], which indicates the proportion of total gained Fairness of Auction Mechanisms — Fairness utility that each player in the coalition shares, ensures that all buyers (or sellers) are treated the utility obtained by the virtual bidder is allo- equally and at least receive (or sell) some basic cated to each of the bidders inside the virtual amount of commodities. However, compared to bidder group. efficiency, fairness is a relatively abstract metric. Reusing the spectrum may not be perfect, Different kinds of fairness measurements can be and collision can occur (e.g., when two cognitive designed to satisfy auction requirements and radio users use the same channel). Therefore, objectives. the collision in spectrum reuse has to be taken IEEE Communications Magazine • November 2012 113C qM IEEE M ommunications q qM Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page MqM q Qmags THE WORLD’S NEWSSTAND®
  7. 7. C qM IEEE M ommunications q qM Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page MqM q Qmags THE WORLD’S NEWSSTAND® mation of the buyers’ strategies (e.g., the bids) and past auction results. Then the belief of the Cognitive radio Trading procedure Trading procedure resource auction of the auctioneer of the bidder seller is calculated from the collected informa- tion by means of the belief function. Based on Spectrum Start Start the belief value, the seller maintains a reserve exploration price for the subchannel it sells. If the incoming Announce auction bid price from a secondary user (i.e., buyer) is for available Wait for No Auction spectrum auction signal? lower than that reserve price, the seller will announcement assert that the buyer is collusive. This is based Collect bids Yes on the intuitive knowledge that the price will Spectrum from bidders Make bid trading decisions most probably drop if there is collusion among Compute auction the secondary users. Likewise, the secondary results and Wait for resource users, as buyers, can also have their belief func- allocate resource allocation tions to prevent primary networks from collud- Learning and Learning and ing as sellers and forming price cartels. Spectrum updating process updating process The belief function approach in [21] still has exploitation Stop Stop shortcomings, since the belief and reserve price need to be generated from at least some non- collusive bids as objects of reference. That is, to set the reserve price as a threshold, the primary Figure 4. Procedure for applying an auction to allocate radio resources in a user has to receive some bids from secondary cognitive radio system. users who are not collusive. In another word, if all the secondary users keep making collusive bids all the time, the primary user may not be into account in the optimization problem formu- aware of the collusion simply because that pri- lated for an auction. The collisions can be mary user even has no idea about what kind of described by using a conflict graph [13], where bid is not collusive. cognitive radio users are defined as nodes, and Besides collusion, other forms of fraudulent adjacent nodes connected by an edge interfere activities may also exist. The sellers (or auction- with each other. eers) can also deceive buyers by overcharging. For example, in a Vickrey auction, the auction- Collusion and Security Issues — Generally, it eer breaks the rule by declaring a payment price is not reasonable to assume that any entity in a higher than the second highest bid price. This cognitive radio environment can either know can be done if the bidders do not know the bid complete information in the system, or have full prices of each other (i.e., not sure about the sec- control over each other. Also, entities in an auc- ond highest bid price). An approach to encrypt tion are self-interested. As a result, in a spectrum bidding language is presented in [22]. In this auction without a strict authority who regulates approach, the actual bid prices from secondary the market, the primary or secondary users may users are encrypted in such a way that the pri- collude to cheat by releasing false information. mary users are unable to know the value of the Such an action is referred to as collusion. bids. The primary users can only compare them. It is pointed out in [7] that collusion is one of As a result, the second highest bid price in the the critical problems that affects the perfor- winning bid cannot be changed arbitrarily by the mance of auctions. Some auction mechanisms primary user, successfully avoiding cheating. are naturally vulnerable to collusion. Reference [20] proves that in a VCG auction, buyers may System-Level Design — To implement the collude to make certain bids, abusing a weakness radio resource auction, the system-level design of the VCG mechanism that buyers can occa- issues need to be addressed, including working sionally pay zero money to obtain commodities. flows of auctioneer and seller/buyer components, Likewise, the sellers (e.g., spectrum owners) can auction system model, and design of auction collude to form a cartel and keep the price high. messages. Therefore, it is important to detect and avoid As shown in Fig. 4, in a cognitive radio sys- collusion. tem, an auction is applied as a spectrum trading It is difficult for the cognitive radio users to process between spectrum exploration and spec- know whether collusion exists or not since they trum exploitation. The spectrum trading process only have limited knowledge. In addition, there consists of workflows of two key components: might be no centralized market regulator (e.g., the auctioneer and the bidder. As shown in Fig. the auctioneer) who can always monitor the sys- 4, during each auction period, the auctioneer tem. Therefore, each cognitive radio user has to announces the auction to share the available build the knowledge of other users as a way to spectrum, collects bids from buyers, and then successfully detect collusion behaviors. Refer- computes the optimal auction results and allo- ence [21] proposes an expression of a belief cates resources to corresponding secondary function to describe the belief of primary and users. The bidder waits for an auction announce- secondary users in the other users’ possible auc- ment, and then decides whether or not to bid. If tion strategies. The belief function is only based the bidder decides to bid, the bidder submits the on the received information and local observa- bid and waits for auction results. Moreover, after tion, but can be used to estimate the behavior of the auction results are revealed at the end of other users. Reference [21] uses the belief func- each auction period, both the auctioneer and the tion to prevent collusion behaviors as follows. A bidder may update their knowledge about the primary user (i.e., seller) keeps collecting infor- auction. 114 IEEE Communications Magazine • November 2012C qM IEEE M ommunications q qM Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page MqM q Qmags THE WORLD’S NEWSSTAND®
  8. 8. C qM IEEE M ommunications q qM Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page MqM q Qmags THE WORLD’S NEWSSTAND® Figure 5 simplifies the system model in [23] and shows the radio resource auction from a Auctioneer Bidder medium access control (MAC) perspective. The auction (or seller) agent and bid agent run the auction processes on behalf of the auctioneer Auction DL DL UL UL Bid agent MAC layer and bidder, respectively. The auction agent and agent bid agent interact with the scheduler in the MAC layer to determine the channel access poli- cy, which subsequently sets the physical layer Queue scheduler Queue scheduler transmission parameters. The auction and bid agents interact with each other through the auc- tion message exchanged in the MAC layer. The general frame structure [24] of an auction mes- Physical layer sage is shown in Fig. 6. The length of each frame depends on the amount of bidding information of an auction process. The header of the frame Figure 5. Auction agent and bid agent. includes control information (e.g., the number of bids included in the frame and priority flag). The bidding signal or bidding message body may include the number of commodities, the bid A frame of bid: price, and other information. Header Bidding signal SURVEY ON AUCTION-BASED DESIGN APPROACHES Number of Bid price Other bidding commodities information... In this section, we review different auction mechanisms used in the literature for the design of cognitive radio systems. Figure 6. An example of a bid frame design. SINGLE-SIDED AUCTION AND DOUBLE-SIDED AUCTION and de-allocation are fixed. In an asynchronous auction, the bidding and allocation processes are Depending on the system model and assump- done at different times. The analysis in [25] tion, an auction can be single-sided or double- shows that in a multiple-commodity scenario, the sided. If there is only a single primary user or synchronous auction performs better than the single secondary user, a single-sided auction can asynchronous auction in terms of allocation effi- be applied (i.e., forward or reverse auctions). On ciency. That is because the receiving and clearing the other hand, if there are multiple primary processes of an asynchronous auction is compu- users and secondary users, a double-sided auc- tationally intensive in the algorithm. tion is applied. The single-sided auction can be extended to A simple scenario of a single-sided auction the double-sided auction [26, 27] when there are for the cognitive radio system is discussed in multiple spectrum sellers (i.e., primary users) in [25], where every secondary user (i.e., buyer) can the cognitive radio system. The primary users access only one subchannel during one auction can compete with each other to attract buyers period. Sequential bidding and concurrent bid- (i.e., secondary users) to buy their available ding are discussed. In sequential bidding, every radio resources. In [27], some side effects are subchannel is taken out by the auctioneer and discussed when the single-sided auction is sequentially auctioned to the buyers, until all the extended to the double-sided auction. In particu- subchannels are sold out. In concurrent bidding, lar, the traditional VCG mechanism is extended the buyers make bids together at the same time to the double-sided VCG auction by considering for the subchannels they request. According to multiple sellers. One finding is that if there is no the definitions, sequential bidding outperforms modification to the auction rules, the revenue of concurrent bidding in the proposed model in the auctioneer can become negative after auc- terms of social welfare metrics, because every tion (i.e., violating ex post budget balance). A subchannel can surely be sold in the end. How- simple example is that, according to VCG’s “sec- ever, concurrent bidding causes less communica- ondary-price” payment rule, each winning bidder tion overhead since each buyer just bids only in the double-sided VCG auction actually pays a once. But the maximum number of total bids in lower price than its submitted bid. Therefore, a sequential auction could be m + (m – 1) + … the final payments are actually lower than the + (m – n + 1), supposing that there are n buy- clearing price (i.e., supply and demand equilibri- ers competing for m subchannels. um). This brings the possibility that the total Reference [25] then analyzes an auction in a payments are lower than sellers’ total ask prices, multiple-auction commodity scenario in which and causes a negative balance of the auctioneer. each secondary user can bid for more than one In a double auction, after the auctioneer col- subchannel. The synchronous and asynchronous lects asks and bids from primary and secondary auctions are considered. The secondary users users, respectively, the auctioneer will perform a make bids together and are allocated with the matching algorithm to clear the market. The subchannel at the same time in a synchronous matching algorithm is based on the spectrum auction. The time intervals between allocation supply (i.e., sorted ask prices from primary IEEE Communications Magazine • November 2012 115C qM IEEE M ommunications q qM Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page MqM q Qmags THE WORLD’S NEWSSTAND®
  9. 9. C qM IEEE M ommunications q qM Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page MqM q Qmags THE WORLD’S NEWSSTAND® Price Price Bid 1 Bid 2 Bid Total Total supply supply Bid 3 Ask 6 curve curve Pb4 Bid 4 Ask 5 Pe Pa4 Ask 4 Bid 5 Pa2 Ask 3 Ask 2 Total Ask Total Pa1 Bid 6 Ask 1 demand demand curve curve 10 22 ... Demand-supply Amount Demand-supply Amount equilibrium equilibrium (a) (b) Figure 7. a) Discrete supply and demand curves of a double auction; (b) continuous supply and demand curves of a double auction. users) and demand (i.e., sorted bid prices from tion model may not be applicable. In this case, a secondary users). An example of spectrum sup- distributed auction model would be required. ply and demand is shown in Fig. 7a, which is a A centralized auction is a straightforward discretized version of the standard supply and approach. For example, [9] discusses a typical demand curves used in economics (e.g., Fig. 7b). scenario, where a centralized base station and The x-axis of Figs. 7a and 7b is the ask/bid multiple cognitive radio users exist in a single amounts, and the y-axis is the ask/bid prices. cell. A single-commodity Vickrey auction is Taking the supply as an example, one primary applied in which a time slot is auctioned during user asks to sell 10 subchannels at price Pa1 (i.e., each auction period. The cognitive radio user Ask 1), which is the lowest price. Another pri- winning this auction can exclusively use the time mary user asks to sell 12 subchannels at price slot for data transmission. A non-cooperative P a2 (i.e., Ask 2), which is the second lowest game is used to analyze this centralized auction price, and so on. with two cognitive radio users as the players. To clear the market, the demand-supply equi- Each user has its own bidding strategies in librium [3], or competitive equilibrium, is deter- response to the other’s possible strategies. The mined from an intersection between supply and user aims to maximize its own throughput, and demand. In a double auction, there can be more at the same time make the other user pay as than one demand-supply equilibrium if the sup- much as possible. Given the second-price auc- ply and the demand are step functions. For tion rules and a user’s purchasing power, the example, as shown in Fig. 7a, the equilibria can Nash equilibrium of bid prices made by users is be any price between P a4 and P b4. However, if proved to exist but its uniqueness is not guaran- supply and demand are continuous, there will be teed. The Nash equilibrium strategy will lead to only one equilibrium, as shown in Fig. 7b. This a unique resource allocation outcome if the demand-supply equilibrium is often used as the channel state distribution is continuous and clearing price. If there are multiple possible starts from zero (e.g., a uniform distribution clearing prices, a uniform price can be employed over [0,1]). In addition, a centralized scheduler to determine sellers’ and buyers’ payments. The can be designed to completely control the auc- auctioneer can choose a single uniform payment tion process. The scheduler collects information price, at which both sellers and buyers must from cognitive radio users and allocates the time make deals. The auctioneer can also choose two slot according to Nash equilibrium strategy. distinctive uniform payment prices for sellers In a system with multiple primary base sta- and buyers, respectively, and earn bid-ask tions with overlapped transmission coverage, spreads. In [28], such a mechanism is used in cognitive radio users have multiple choices of spectrum trading of primary and secondary users choosing service from the base stations that are in the cognitive radio system. the radio resource sellers. Such a process could be decentralized. In such a situation, the central- CENTRALIZED PROCESS AND ized auction may not be suitable. Therefore, [28] DISTRIBUTED PROCESS introduces a distributed double auction mecha- nism, called the multi-auctioneer progressive Traditionally, auctions are held in a centralized auction (MAP). In this mechanism, cognitive fashion. In particular, buyers and sellers submit radio users (i.e., buyers) interact directly with bids and asks to the central auctioneer, respec- primary base stations (i.e., sellers) without a cen- tively. The auctioneer uses bids and asks to clear tralized controller. The seller starts selling a the market given the auction rule and regula- channel by asking a low price of all users. The tion. However, some cognitive radio systems may ask price increases periodically. The cognitive not have a centralized entity to manage the spec- radio users decide whether to buy a channel and trum allocation, and hence the centralized auc- from which seller. The auction ends when the 116 IEEE Communications Magazine • November 2012C qM IEEE M ommunications q qM Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page MqM q Qmags THE WORLD’S NEWSSTAND®

×