Background of the Invention Delivering information via wireless transmission is becoming increasingly popular.

As seen in FIG. 1, information, such as stock prices, traffic information, weather reports, airline schedules, and sports scores, may be broadcast from a single source, such as a service provider 30 (not unlike a cellular phone service provider) to a number of recipients (users) 32 via a wireless transmission media. These media may be, for example, paging networks, FM subcarrier networks, cellular phone networks, and PCS

(personal communications services) networks. This application will refer to the media simply as a "wireless media network".

A user 32 may be a person having a wireless terminal 34, such as a personal digital assistant (PDA), as is illustrated in FIG. 1. (The wireless terminal 34 is often referred to as a "client" , and the ultimate human recipient of the information is referred to as a "user".) A PDA is typically a laptop or palmtop computer connected to a wireless media network. The wireless terminal 34 usually is not connected to any direct power source, but rather runs on either conventional or rechargeable batteries. Because wireless terminals are often used away from the home or office, it is an important consideration of a wireless terminal user to maximize the length of time that the terminal may operate without having to change or recharge batteries. Thus, it is important to minimize the power consumption necessary for the client to receive, decode, and display information received over the wireless data network, thus increasing the useful life of the battery. One method of transmitting information over a wireless network is to broadcast the information periodically. This method is well-known and has been discussed in Imielinski et al, "Energy Efficient Indexing on Air", Proc. ACM SIGMOD Conference, May, 1994. In August 1994 the inventors noted the similarity of this method to the

method of writing data on a standard rotating magnetic disk, and thus called the method an "airdisk" by analogy. Subsequently, Zdonik et al, "Are 'Disks in the Air' Just Pie in the Sky?", Proc. IEEE Workshop on Mobile Computing Systems and Applications, Dec. 8-9, 1994, also noted this similarity, and called this method a "broadcast disk" . An "airdisk" is a periodic transmission of data over a wireless network. It is called an airdisk because it may be theoretically compared to a rotating data disk, as will be illustrated below. FIG. 2 shows a periodic broadcast 40 having a number of transmissions (the arrow indicates increasing time). A first transmission 50 includes the following sequence of topics: stock prices 51, traffic 52, weather 53, airline schedules 54, and sports scores 55. After the transmission 50 is complete, a new transmission 60 is immediately sent, beginning with, in this example, stock information 61, traffic 62, etc. Each transmission may begin with a preamble 56, 66 indicating the beginning of a transmission. The preamble may be followed by an index 57, 67 which indicates the location of the beginning of each topic in the transmission. Each topic 51 - 55, 61 - 65 may begin with a topic header 58, 68 which indicates the start of a topic. Each transmission may end with a trailer 59, 69 indicating the end of a transmission.

Although the example in FIG. 2 shows the broadcast is composed of different items defined as topics of information (stock prices 51, traffic 52, etc.), it could just as well be the case that the broadcast contains items which relate to a single topic. For example, FIG. 2 could refer to stock prices only, with a first transmission including the following sequence of items which relate to a single topic: IBM price, NYNEX price, HP price, and so on. This discussion often refers to "topics" for illustration. In most instances, however, the illustration is equally applicable to smaller items, such as information about particular stocks, traffic conditions on certain roadways, or weather conditions for a certain geographic region.

The broadcast 40 is periodic because each transmission is immediately followed by another. As seen in FIG. 3, the periodic broadcast 40 of FIG. 2 may be theoretically compared to a revolving disk 70 (i.e., such as a computer or optical disk) and a read head 80 (i.e., such as a magnetic or optical head) (the arrow indicates the direction of disk rotation). The disk in this illustration is separated into five radial portions called sectors. These sectors contain stock 71, traffic 72, weatiier 73, airline 74, and sports 75 information, respectively. Protocol related portions of the signal such as the preamble,

the index, the headers, and the trailer are omitted for simplicity. After a complete rotation of the disk 70, the same sector is presented to the head 80 for reading. Similar to a computer disk, die periodic broadcast 40 may be updated by the service provider 30 so that subsequent rotations may include revised, additional, or altered data; and the data may be presented in a different sequence.

Much of the information on the airdisk is dynamic - for example, stock prices, sports scores, weather, and traffic conditions may change throughout the day. Thus, after a period of time some information transmitted in a periodic broadcast may become "stale" and is of little use to the user 32 (i.e., hours old stock prices during active trading). Thus, the amount of time it takes a client 34 to access all of the desired information (referred to as "access latency") is one measure of performance for an airdisk transmission. Also, because many periodic broadcast clients may be wireless terminals, where power efficiency is a major concern, power consumption may be minimized by the ordering of the data in the transmission. This is illustrated in FIGs. 4a and 4b. Using the rotating disk analogy, FIGs. 4a and 4b show the importance of data ordering in instances where the transmission is not indexed and indexed, respectively. Using an index in a periodic transmission is discussed in the Imielinski et al. paper cited above.

Where the transmission has no index, d e receiver, such as a wireless terminal 34, must be on at all times to determine whether it is receiving information it is interested in so that it may read this information. This constant monitoring of the incoming transmission inefficiently consumes power. Alternatively, if all of the information sections are of equal size ~ and thus of equal time length - the terminal 34 could be programmed to turn on at die beginning of each section and quickly determine if the section includes data desired to be read. This alternative conserves power but may be impractical because the sections may be of differing lengths and because d e information is dynamic, each section may be a different length each broadcast. (Unlike a physical disk, an airdisk may increase or decrease in size if iiiformation is added or deleted to a topic.) If die sections are certain types of data items, such as stock prices, however, the items may be arranged to have identical lengths.

An index may be provided at die beginning of each transmission giving the sequence of topics and die location of the beginning of each topic in the upcoming transmission. This is advantageous because it allows die terminal 34 to be "off" (i.e., consuming a reduced amount of power) except when desired information is being broadcast. This reduces e "on" time for the wireless transmitter 34 and conserves power. An index may be disadvantageous, however, because it requires additional data to be included in d e broadcast. This makes me airdisk "larger" and takes a longer time to transmit the entire periodic broadcast.

FIG. 4a illustrates the access latency for an unindexed airdisk 70'. A user 32 is interested in sports and traffic information. At the random time shown in mis figure, me read head 80 is positioned over airline information 74' . A first time delay - a rotational latency 82

— occurs in die time it takes the disk to rotate to a topic of interest (in this illustration, sports is the first topic of interest presented to die read head 80). The next time period is die read time 84 (illustrated by a dashed line), that is, the time that the information about die desired topic is being read from the disk 70' by d e read head 80. A second time delay - a topic (or item) spread 86 - occurs during me time it takes d e disk to rotate to the last topic of interest. The final time period is d e final topic read time 88 when the final information about me topic of interest is read by die read head 80. Because information is read on- ie-fly (i.e., it is read relatively instantaneously as it rotates past me read head 80), die total access latency would be the same if some or all of die topics between sports and traffic were also desired. That is, in this illustration, d e access latency and die terminal "on" time would have been the same if the user wanted stock information 71', as well. Thus, in d is example, the access latency for all of die desired topics is about tiiree quarters of a rotation (about three quarters of a complete transmission).

FIG. 4b illustrates the access latency for an indexed airdisk 70" . Note that the airdisk 70" includes an index 90. Again, me topics die user 32 is interested in are sports 75" and traffic 72". At die random time shown in this figure, die read head 80 is positioned over airline information 74". A first time delay - a rotational latency 92 for the index 90 - occurs in d e time it takes me disk to rotate to the index 90. Note that the sports information 75", which is die user is interested in, was not read because the

index 90 has not yet been read and dierefore the client 34 does not know die location of the sports information. Thus, the sports information 75" is passed over until die next rotation (i.e., me next transmission). The next time period is die index read time 94 (illustrated by a dashed line). This is the time mat d e index 90 is being read by d e read head 80. A second time delay ~ a topic (or item) reach 96 ~ occurs during the time me disk rotates to die last topic of interest (here, sports 75"). Note that during the topic reach 96, information about topics of interest (traffic 72") is read. The traffic information read time 97 (illustrated by a dashed line) is part of the topic reach 96. Because reading is done on-the-fly, this read time 97 does not affect the topic reach 96. The final time period of the access latency is d e final topic read time 98 when information about die final topic of interest (sports 75") is read by d e read head 80. Because of die index 90, the terminal 43 is only "on" when information about topics of interest is being read. The total "on" time for the terminal 34 is die index read time 94, the traffic information read time 97, and die sports information read time 98. Thus, the total on-time is reduced from that of die non-indexed airdisk of FIG. 4a. On the other hand, die total access latency is much greater in this illustration, and exceeds die time for a complete rotation due to the index rotational access latency 92.

"Access latency" as used in this patent application, refers to the total time for: a) e rotational latency; b) die first item read time; c) the item reach or spread (including any read times); and d) die last item read time.

Actual data disks have anodier time delay as illustrated in FIG. 4c. A disk 70'" may be arranged with data stored in sectors divided by annular tracks. In FIG. 4c, the disk 70'" is divided into five sectors and three tracks. This creates fifteen areas for storing data (i.e., data 0 - data 14). The additional time delay is due to positioning the read head 80 over d e appropriate track when the sector is positioned by d e read head 80. This time delay does not apply to die disks 70', 70" because diey have only one track, nor to airdisks because there is no head to be physically positioned over a track.

One way to reduce d e average access latency for a number of users 32 would be to order die topics on d e transmission so that die average access latency for all users 32 is minimized. The rotational latency and data transfer times are fixed once die set of information in a given rotation (transmission) is decided. However, die topic spread (or topic reach) may still be varied by knowing in advance die sequence of the topics on d e disk. If die topics which d e users 32 desire to read is known in advance (how this may be done is described below), a disk layout (i.e., a transmission) may be chosen to reduce die average topic spread or reach for all users. That is, if many users are interested in stock and traffic information, d ese users' access latency may be minimized by placing these two topics first.

The inventors of the present invention, however, have proven diat die algorithm minimizing me access latency for a number of users 32 each requesting two or more items is "NP-complete" (non deterministic polynomial time complete). This means diat d e algorithm is mathematically intractable to calculate. This is discussed in Jain and Werth, "Airdisks and AirRAID: Modeling and Scheduling Periodic Wireless Data

Broadcast (Extended Abstract)", presented on April 27, 1995 at Rutgers University Winlab Workshop. The contents of this paper are incorporated herein by reference.

Anodier problem associated with periodic wireless data broadcasts is d at an increased data amount results in a "bigger" airdisk. That is, die time for a complete periodic transmission increases. This longer transmission time results in a longer potential access latency which, as discussed above, is undesirable due to d e dynamic nature of some of the information in die broadcast. Also, in many existing wireless spectrum transmission channels, data communications must coexist with voice communications. The wireless spectrum is divided into multiple channels, where the bandwiddi is optimized for voice or short message communications (such as paging).

For example, a Wireless Access Communications Systems (WACS) has been proposed by Bellcore in a paper entided "Generic Criteria for Version O.I. Wireless Access Communications Systems (WACS)" , Technical Advisory TA-NWT-001313, Issue 1, July 1992. This system uses transmissions having frames divided into time slots.

Each frame has eight time slots. Each time slot is 2.5 msec and is capable of holding 32 kb/sec. (Other wireless transmission media may be capable of broadcasting at only

16 kb/sec.) Thus, in one minute a broadcast could uansmit 1.92 Mbits. However, data "overhead", such as packet framing, indexes, preambles, headers, trailers, and die like reduce die amount of actual data in a transmission, effectively reducing die speed of a transmission. Thus, a 1 megabyte (i.e., 8 megabit) transmission may take as long as eight minutes. Moreover, as the amount of data in d e transmission increases, die transmission time (me rotation time) increases linearly. The increased transmission time (the rotation time) results in an increased access latency. Because d e value of some information decreases with time (i.e., stock prices during active trading), a high access latency is unacceptable. Thus, it is desirable to increase d e "storage capacity" of an airdisk, so diat a large number of bytes are transmitted per unit time.

Thus, it is an object of die present invention to provide a method for arranging the information on a periodic transmission which reduces die average access latency for clients of a wireless data service provider.

It is anodier object of the present invention to provide a transmission signal structure which increases an airdisk' s storage capacity without incurring undesirable access latency penalties.

It is yet anodier object of die present invention to provide an efficient periodic wireless transmission having adequate error correction.

Summary of the Invention

These objects are achieved by a periodic wireless data transmission according to die present invention. The order of topics or items on an airdisk may be arranged by sorting d e topics or items to be transmitted onto a list in order of popularity. In one preferred embodiment, this list may be revised by comparing each adjacent pair of topics on die list. The topics' positions on the list are exchanged if d e exchange decreases die average latency for all users. This is repeated for all of die topics on die list.

In a second preferred embodiment, users are provided with a number of votes to select items or topics of interest, which may be "cast" in any manner the user chooses (even casting fractions of one vote). This prevents users interested in a few items or topics from being "outvoted" by users interested in many items or topics. One way tiiis may be done, for example, is by providing each user widi one "popularity vote", which may be used on one or more items. If more than one item is chosen, the value of the

user's vote is decreased to d e fraction 1/x, where x is die total number of items the user selected. The value of all d e votes for an item or topics is added in determining the order of popularity.

In a third preferred embodiment, after die popularity list is sorted, die comparison of adjacent topics or items is performed iteratively. For example, once die most popular topic is determined, d e process is repeated for all users not interested in d e most popular topic or item. This is repeated until all of d e topics or items have been sorted.

The transmission structure may also be arranged so d at die transmission combines a number of transmission channels to obtain greater aggregate capacity. This may be done, for example, by structuring the transmissions to be theoretically comparable to a plurality of "striped" data disks. Data disk striping is writing adjacent data units across an array of synchronized rotating disks. This allows data to be accessed in parallel. In a preferred embodiment, data is "striped" onto a wireless media transmission by writing adjacent data units across a number of frequencies of an FDMA transmission. This allows a client to receive d e information of interest simultaneously from a plurality of frequencies.

Brief Description of the Drawings

The present invention is described widi reference to the following figures: FIG. 1 illustrates a wireless data service;

FIG. 2 illustrates a periodic wireless data transmission;

FIG. 3 illustrates an "airdisk" analogy to d e wireless data transmission illustrated in FIG.

2;

FIG. 4a illustrates access latency in an unindexed airdisk; FIG. 4b illustrates access latency in an indexed airdisk;

FIG. 4c illustrates a data disk having several tracks;

FIG. 5 is a block diagram of a first system used to obtain user interest information;

FIG. 6 is a block diagram of a second system used to obtain user interest information;

FIG . 7 is a flow chart of one metiiod for arranging information on a periodic transmission according to d e present invention;

FIG. 8 is block diagram of a system for arranging information on a periodic transmission;

FIG. 9 is a user interest table;

FIG. 10 is a sorted list based on die user interest table of FIG. 9;

FIG. 11 is a calculation of the objective function value of die sorted list of FIG. 10;

FIG. 12 is a calculation of die objective function value of a first revised sorted list; FIG. 13 is a calculation of die objective function value of a second revised sorted list;

FIG. 14 is a calculation of die objective function value of a tiiird revised sorted list;

FIG. 15 illustrates data striping on an array of data disks;

FIG. 16 illustrates seven levels of RAID error correction;

FIG. 17 compares an FDMA signal according to one embodiment of die present invention with an array of data striped disks; and

FIGs. 18a, 18b, 18c, 18d, 18e, 18f, and 18g illustrate seven levels of error correction for transmissions according to die present invention.

Detailed Description of a Preferred Embodiment A. Arranging Information On A

Periodic Wireless Data Broadcast

To arrange topics on a periodic wireless data transmission in a manner which reduces d e average access latency for all users, the service provider 30 must have available some indication of which topics d e users 32 have an interest. The service provider 30 may be made aware of these interests in any number of ways. Three of these methods are described below.

A first method for a service provider to obtain client interest information is to implement a profile-based system. Many communications systems, for example cellular telephone services, include a database, or "profile", of client information. The customer profile for a cellular telephone service provider may include, for example, data such as die user's name, address, long distance telephone service provider, service features (e.g. , call forwarding and call restriction), billing, and other administrative related information.

FIG. 5 shows a profile based wireless data service 110. In a profile-based information service, a user service profile may be used by die service provider 30 to obtain user interest information. One example may be a service provider 30 diat provides traffic and weather information for several geographic areas, and New York Stock

Exchange ("NYSE") listings. When a user 32 subscribes to die service, a profile 112 for diat user 32 may be created either by die service provider 30 using a computer or die like or by die user 34 over a wireless media network 114 via a wireless terminal 34, over a telephone network 116 via a telephone 118, or otiier communication device. The profile 112 may indicate, for example, diat between 6:00 a.m. - 7:00 a.m. the user 32 is interested in weadier conditions for die user's geographic region, d at between 7.30 a.m. - 9:30 a.m. and 5:00 p.m.- 6:30 p.m. the user is interested in traffic information about certain highways and bridges over which die user commutes to and from die office, and diat between 9:30 a.m. - 3:30 p.m. die user is interested in information about several stocks on the NYSE. The user 32 may occasionally update die profile, for example to receive traffic information for a route to a clients' office, by communicating mis interest to die service provider 30 via die wireless terminal 34, telephone 118, or other communication device. By obtaining user profile information, the service provider 30 knows sufficiently in advance d e topics and/or items each client is interested in at certain periods throughout the day. A broadcast assembler, or airdisk controller 120, uses the profile information for all or some of die users to arrange die topics and items and assemble the transmission in a manner described below. The assembled transmission is sent to a wireless media transmitter 122 for broadcast.

A second method for d e service provider 30 to obtain client interest information is to directly poll a cross-section of all of die service provider's customers. This will provide a "Nielsen ratings"-type profile of die customers' information interests. The information provider 30 may also use its knowledge of the particular needs of its customers to arrange information to help reduce die average topic (or item) spread or reach, thus reducing die average access latency for all clients. For example, a service provider broadcasting traffic mformation to users commuting between New York City and New Jersey might reasonably guess diat consecutively ordering traffic information about conditions on die Holland Tunnel, Lincoln Tunnel, and die George Washington Bridge will reduce die average topic spread or reach.

A third, dynamic system to obtain client interest information is shown in FIG. 6. This system 130 provides the wireless terminal 34 widi an uplink to die service provider

30. An uplink is an active communication channel originating from the client 34 to the service provider 30. The uplink is illustrated in FIG. 6 by the two way arrow between

die client 34 and die wireless media network 114. Users 32 can request information as it is needed by sending a request over d e uplink. In an illustrative embodiment of this metiiod, die service provider 30 monitors die uplink for a period of time referred to as die "batch length". The provider counts die number of requests received for each data item during die batch lengtii period. Aldiough die batch lengtii may be any length, in a preferred embodiment me batch length is just less tiian die lengtii of die current rotation (i.e., the complete length of the current transmission). The collected requests are used to arrange the topics on the next transmission.

It may be preferable for the service provider 30 to arrange die topics on die upcoming transmission based on only a cross-section of die requests received during die batch length. It may also be preferable to provide a method where users requesting unpopular information are not "starved". One method for achieving tiiis end is as follows: a service provider 30 providing T total topics for broadcast includes in each rotation a subset of d e p most popular topics (where p < T). The broadcast would then include transmissions which dynamically vary subsets of die remaining, less popular topics. This may be done, for example, by including in all broadcasts d e most popular topics or items as determined by an average over time. Less popular requested items would be broadcast in a subset of all transmissions. If tiiis metiiod is used wid die user uplink system of FIG. 6, d e service provider 30 may disregard requests for the most popular items because it has already been determined diat they will be included in die next transmission. The service provider may men process requests for less popular topics or items, ti us reducing die number of requests needing processing. Alternatively, users may be precluded (eitiier permanently or for a limited time period) from issuing requests for die most popular items, thus having d e further benefit of conserving uplink bandwiddi.

FIG. 7 is a flow chart of a metiiod 200 for arranging topics on a transmission according to an embodiment of die present invention. In tiiis illustrative method, the user interest information is collected by die service provider 30 via user profiles as seen in FIG. 5. The airdisk controller 120 and transmitter 122 may be conventional wireless media transmitting equipment programmed to perform die following method, as seen in

FIG. 8. The airdisk controller 120 comprises a computer 150 and a data processor 152, such as a digital signal processor. The user profile information 112 includes a user

profile database 154. The computer 150 receives die data to be included from a data input 156, which may be a manual input, a wireless or wireline network, or otiier means for providing information to me computer 150. The computer 150 also retrieves user interest information from the database 154 to arrange die topics and items into a desired order. The arranged topics and items are then sent to d e data processor 152 to be converted into a format suitable for transmission. The convened data is then sent to a conventional wireless media transmitter 122. As noted above, die user may be interested in specific items witiiin topics. That is, the user may be interested only in certain stock prices or traffic information on certain roads and bridges. First, die airdisk controller 120 is initialized (step 202). The user interest information, such as the user profiles, is accessed and read (step 204). The airdisk controller 120 creates or updates a user interest table 250, illustrated in FIG. 9 (step

208). The user interest table 250 may be compiled from die accessed user interest information. The number of rows in the table equals the number of users interested in one or more topics at the current time. In die example of FIG. 9, at me current time, nine users are interested in receiving information. The columns in the table represent topics. Using the examples given in FIGS. 2-4, tiiis user interest table 250 has five columns (stocks, traffic, airline schedules, weather, and sports). For each row . (i.e., for each user), a value, such as a "1" is assigned to a column j if the user i is interested in topic or item ./, and a "0" otiierwise. If a column has no Is in it, that column is deleted. At this time, no user is interested in airline information, so this column is deleted. The total number of columns in the final table is n. In this example, n = 4.

Next, a list K is calculated (step 208). The value of K(j), theyth topic on die list, is die total value (i.e., die number of "1" entries in column j of die user interest table) assigned to that topic. List K is sorted in ascending order of Kφ values (step 210). The sorted list is called L. FIG. 10 is die list L for die user interest table 250 of FIG. 9. Of the topics remaining on die list, stock information was die most requested topic (6 requests) and sports information was die least requested (3 requests)(airline information had no requests and was deleted from the list). L is the th topic on list L. j ranges from 1 to π.

Next, M, die value of the objective function for list L is calculated (step 212). M is die sum of die maximum position value of die topics a user is interested in when die topics are ordered according to the list L for all users. FIG. 11 illustrates the calculation of M for die list L of FIG. 10. As an approximation of a method to minimize die access latency for all users 32 (i.e. , users 1 - 9), die topics are arranged on die list from die most to the least popular. The average access latency for all of the users obtained by adding up numerical values for the positions of the last topic of interest for each user. The numerical value for the positions of die topics on list L are:

Stocks 1

Traffic 2

Weather 3

Sports 4

Thus, user l 's last topic of interest is sports (4); user 2's last topic of interest is weather (3). The numerical values for all nine maximum positions are added up to determine M for list L. Here, M is 26.

Lety = 1 (step 214) and determine ϊj = n (step 216). If/ is not equal to n, men steps 218 - 226 are preformed. If j equals n, tiien steps 228 and 230 are performed.

If j is not equal n, then the positions of topics Lφ and L(j + 1) are interchanged and a new list L ' is created (step 218a). In this example, die positions of weather and sports are exchanged. The order of new list L, ' is:

Stocks 1

Traffic 2

Sports 3

Weadier 4 M die value of die objective function for list new list L,' is calculated (step 218b) in die same manner as above. The calculation for M, ' is illustrated in FIG. 12. Then M and M' are compared to determine if M' is less than M. If M ' is less tiian M, d en die new list I' replaces the original list L, and die original list is discarded (step 222). If M ' is not less than M, tiien die new list L ' is discarded (step 224). Here, M, ' > M. New list L, ' is discarded and sports is returned to the last position on die list. After M' and M are compared, j is increased toy +7 (step 226).

Next, the positions of traffic and weadier are exchanged. The order of new list

L ₂ ' is:

Stocks 1

Weadier 2 Traffic 3

Sports 4

M ₂ the value of the objective function for this new list L ₂ ' is calculated (step 218b) in die same manner as above. The calculation for M ₂ ' is illustrated in FIG. 13. In tiiis case, M ₂ ' < M and dierefore, new list L ₂ becomes list L, M ₂ ' becomes , and the positions of weather and traffic remain exchanged. A final list L ₃ ' is assembled having me positions of weather and stocks exchanged. The order of new list L ₃ ' is:

Weadier 1

Stocks 2

Traffic 3 Sports 4

MY the value of die objective function for this new list L is calculated (step 218b) in die same manner as above. The calculation for M ₃ ' is illustrated in FIG. 14. In this case, M _} ' > M, and this list L is discarded.

This process is repeated until all n (all topics on the table) are sorted according to steps 218 - 226. Once all of d e topics are sorted, a final list L _f (which in this example is list L ₂ ') is provided to die airdisk controller 120 which retrieves and arranges the topics, and generates a transmission which is transmitted by transmitter 122 (step 228).

The airdisk controller 120 waits until anodier broadcast is to be assembled (step 230).

This method is relatively cost effective for the service provider because it may be performed wid conventional equipment. For example, a single computer appropriately programmed may arrange the topics and items for each transmission.

Similar methods may also be used to arrange die information on the periodic wireless data broadcast. One alternative to d e metiiod described above is to modify step

208 by utilizing the user interest table 250 in a way which provides each user with equal "weight" . That is, each user is given a finite number of "votes" (one, for example), and may cast diose votes in any way the user desires, including casting a fraction of one vote.

Thus, a user interested in only one stock, for example, may cast all of his votes for one stock. Anodier user could spread all his votes among several hundred stocks. Thus, the user interested in only a few items or topic is not "outvoted" by a user interested in

several items or topics. A preferred metiiod for accomplishing this end is for each column j, consider each user / which has an interest in the item in that column. Instead of a "1" being placed in die column, a fraction 1/x is placed tiiere, where x is the number of items user i is interested in. After tiiis is completed for all columns and all users, die values for each column (i.e., topic) is summed, and die topics are sorted by descending order of tiiese sums. Then the procedure in steps 212-230 is performed as described above.

Anodier alternative is to use eidier of die two mediods (the metiiod shown in FIG. 7 or me one in die preceding paragraph) and perform diem iteratively. That is, after the first topic in d e final list has been chosen, delete all rows in die user interest table which have a non-zero value for that topic, and repeat the metiiod to determine d e second topic in die final list; continue with this process until all die topics have been placed in die final list. Referring to die user interest table 250 of FIG. 8, after the final list L _f is determined, die entire process is repeated, except diat die users interested in stocks are deleted. That is, die process is repeated for users 3, 8, and 9 only. If tiiis second process determines, for example, diat traffic is the second item on the list, a third process is performed for user 3, who is not interested in traffic or stocks.

B. Structuring An Efficient Periodic Wireless Transmission

The "storage capacity" of an airdisk may be increased by structuring the periodic broadcast to combine several channels to obtain greater aggregate capacity. One way this may be accomplished is by structuring the periodic broadcast to be theoretically compared to d e technique of data striping, as is used in a Redundant Array of Inexpensive Disks

(RAID).

Data striping has been known for magnetic disks since the 1980s and has been used in high performance workstations and super computers. Data striping is described, for example, in Chen et al., "RAID: High Performance, Reliable Secondary Storage", ACM Computing Surveys, Vol. 26, No. 2, June 1994. The contents of this article are incorporated herein by reference. Briefly, data striping is writing adjacent data units across an array of synchronized rotating disks. FIG. 15 shows an array of four disks 380a-d. The data unit (i.e, bit, byte, block etc.) used to stripe is called die striping unit.

The number of disks in die array is called die stripe widώ. In tiiis figure, the striping unit is 1 bit and die striping widtii is four disks. This structure, often referred to as a

RAID allows data to be read in parallel and dierefore is a faster means for obtaining data from large memories than from a single, large disk. Using multiple disks, however, results in decreased reliability. For example, using two disks together decreases the reliability by half because there is double the possibility diat one of die disks will fail. Thus, a number of redundancy and error correction schemes have been developed for use in RAIDs.

FIG. 16 illustrates seven levels of error correction used in RAIDs. These levels are:

Level 0: no error correction or redundancy. Level 1 : "mirrored redundancy" . All of die data is copied twice; if one disk fails, there is an identical disk containing the data. If n disks are available, d e data is striped across n/2 disks, and these disks are replicated. Level 2: "memory-style ECC". Data is striped across a set of disks, and the remaining disks are used to store a Hamming Code error correction code. The disks widi the Hamming Code are called parity disks. The number of parity disks needed is on die order of magnitude of log ₂ n.

Level 3: "bit interleaved parity". Data is bit interleaved across n-1 disks and one disk is used for parity.

Level 4: "block interleaved parity". This is similar to Level 3, except diat die striping unit is a block, not a bit.

Level 5: "block interleaved distributed parity". Data blocks and parity are uniformly interleaved across all n disks.

Level 6: "P + Q redundancy" . This is similar to Level 5, except diat Reed- Solomon codes are used to protect against simultaneous disk failures. FIG. 17 compares an illustrative embodiment of a FDMA wireless periodic transmission 450 having four frequencies f,, f ₂ , f ₃ , f ₄ with a magnetic disk RAID 452 having four disks 453, 454, 455, 456. Wireless media transmissions divide time into fixed-sized periods called "frames". In FIG. 17, the frames are the portions of each frequency shown (one frame is indicated by die bracket 458). A frame is divided into smaller units. Some of tiiese smaller units are protocol-related information such as die preamble, index, headers, and trailers (omitted from FIG. 17 for clarity). Other smaller units containing user information (i.e., voice or data information) are called "bursts" .

All bursts of a transmission have die same size. The four equally sized subdivisions of a burst including the same information in a plurality of frames is called a "time slot". That is, if die first burst of eight consecutive frames includes stock information, this first burst is called a time slot. For simplicity in this discussion, all bursts are referred to as time slots. In FIG. 17, the four equally sized subdivisions of each frame 458 are time slots.

The FDMA signal of FIG. 17 has four frequencies f,, f ₂ , f ₃ , f ₄ , with each frequency having frames divided into four time slots 460 a- d. The second and fourth time slots 460b, 460d are assigned voice information; die first and third slots 460a, 460c contain data. In tiiis illustrative embodiment, die first and second frequencies f,, f ₂ , contain stock information. The third and fourth frequencies f,, f ₄ , contain sports information.

The first time slot 460a of the first frequency f, contains the first byte (i.e. , byte 0, which includes bits 0 - 7) of stock information, and all subsequent even bytes (i.e. , 0, 2, 4...) to its capacity. In this illustrative embodiment, the slot has a lόkbit capacity

(2kbyte), so the highest byte in tiiis slot is 3998. The first slot 460a of die second frequency f ₂ contains d e second byte (i.e., byte 1) and each subsequent odd byte (i.e. 1 , 3, 5...), so d e last byte of this slot is 3999. Thus, these two simultaneously transmitted frames contain the first 4kbytes of stock information in the transmission. In this illustrative embodiment, d e transmission 450 may be theoretically compared with a

RAID having a one byte "striping unit" and a two disk "stripe width" of stock information.

The third slot 460c of die first frequency f, contains the 4001st byte (i.e., byte

4000) and each subsequent even byte (i.e. , bytes 4000, 4002... ) to its capacity. The third slot 460c of the second frequency f ₂ contains the 4002nd byte (byte 4001) and each subsequent odd byte (bytes 4001, 4003...) to its capacity. The sports information may also be compared to a one byte "striping unit" /two disk "stripe width" RAID.

In these transmissions, the client receives the data twice as fast as would be possible in ordinary wireless media transmissions. Also, the stock information is transmitted on two frequencies (f _t , f ₂ ) and die sports information is transmitted on two different frequencies (f ₃ , f ₄ ). If a single user was interested in both topics, a receiver capable of receiving four frequencies would be needed. If a user was only interested in

one of the two topics, a receiver capable of receiving only two frequencies is needed. It may be possible to have "dedicated receivers" . That is, a user 32 may have a terminal 34 dedicated to receiving a subset of all the frequencies of the FDMA signal. For example, a sports-dedicated receiver could be configured to receive only f ₃ and f ₄ . Different air disks may have different rotation lengtiis. For example, if there is twice as much stock information as sports information die stock "disk" will contain twice as much data and therefore be much larger. Because the data rates of the transmissions are the same, it will take twice as long for the stock disk to complete one rotation than for the sports "disk" . It is also possible tiiat different airdisks may have different sized bursts. This is most likely to occur in FDMA airdisks.

FIG. 18 illustrates error correction formats which may be used in embodiments of wireless data broadcasts according to the present invention. These levels are: Level 0: no error correction or redundancy.

FIG. 18(a) shows a TDM A signal 500 having four time periods (i.e., die periods may be time slots or frames) with no redundancy or parity.

Level 1: "mirrored redundancy". All of the data is broadcast twice; if there is a transmission or reception error with d e initial broadcast, there is an identical broadcast containing the data. If n frames are available, die data is broadcast in n/2 frames, and these frames are repeated in subsequent frames. FIG. 18(b) shows a TDMA signal 510, having with four time periods, data 0 - data 3, and four time periods, redundancy 0 - redundancy 3, containing the identical data as in d e first four time periods.

Level 2: "memory-style ECC". Data is broadcast "striped" across a number of disks, and die remaining frames are used to store a Hamming Code error correction code. The frames with the Hamming Code are called parity frames.

The number of parity frames needed is on die order of log ₂ n. FIG 18(c) shows a TDMA signal 520 having four time periods containing data (data 0 - data 3), and tiiree time periods containing Hamming code error correction bits (parity a - parity c). Here, there are four frames of data broadcast, and l +log ₂ 4 frames are needed, dierefore diree parity frames are needed.

Level 3: "bit interleaved parity". Data is bit interleaved across n-1 frames and one frame is used for parity.

FIG. 18(d) shows a TDMA signal 530 having four time periods containing data (data 0 - data 3), and one time period containing bit interleaved parity (parity a). Level 4: "block interleaved parity". This is similar to Level 3, except diat die striping unit is a block, not a bit.

FIG. 18(e) shows a TDMA signal 540 having four time periods containing data blocks (data blocks 00 - 04, 10 - 14, 20 - 24, 30 - 34) and a one time period containing block interleaved parity (parity aO -a4). Level 5: "block interleaved distributed parity" . Data blocks and parity blocks are uniformly interleaved across all n frames.

FIG. 18(f) shows a TDMA signal 550 having five time periods containing interleaved data blocks (data blocks 0 - 19) and parity blocks (parity blocks 0 - 4). The interleave pattern shown in tiiis figure is called d e left symmetric parity placement, which has been found to be a preferred arrangement for distributing parity blocks.

Level 6: "P + Q redundancy". This is similar to Level 5, except that Reed- Solomon codes are used to protect against simultaneous frame failures. FIG. 18(g) shows a TDMA signal 560 having six time periods containing interleaved data blocks (data blocks 0 - 19) and parity blocks (parity blocks OP -

4P, 0Q - 4Q).

The parity writing for die airdisk may be performed in the airdisk controller 120. Parity, or redundancy, may be needed for each transmission (if information has been revised).

This may be performed, for example, by performing an Exclusive-Or operation of the new data widi d e data from die previous transmission.

It is also contemplated that airdisks may be structured using TDMA and CDMA signals as well. For example, each time slot of a TDMA signal may be structured to be an independent airdisk.