2001-11-12
`
`802.16abc-01/61
`
`Project
`
`IEEE 802.16 Broadband Wireless Access Working Group <http://ieee802.org/16>
`
`Title
`
`Fast and Efficient BW Request Mechanism for IEEE 802.16a OFDM PHY
`
`Date
`Submitted
`
`Source(s)
`
`2001-11-08
`
`Jerry Krinock, Manoneet Singh, Mike Paff,
`Vincent Tien, Arvind Lonkar, and Lawrence
`Fung
`Radia Communications, Inc.
`275 N. Mathilda Avenue
`Sunnyvale CA 94305
`
`Re:
`
`BW Request Mechanism
`
`Voice: 408 830 9726 Ext 239
`Fax:
`408-245-0990
`Email: jkrinock@radiacommunications.com
`
`Abstract
`
`This document presents the results of a detailed simulation and analytical study of various bandwidth
`request mechanisms for the IEEE 802.16a OFDM PHY.
`
`Purpose
`
`Improving current 802.16a standard
`
`Notice
`
`Release
`
`Patent Policy
`and
`Procedures
`
`This document has been prepared to assist IEEE 802.16. It is offered as a basis for discussion and is not
`binding on the contributing individual(s) or organization(s). The material in this document is subject to
`change in form and content after further study. The contributor(s) reserve(s) the right to add, amend or
`withdraw material contained herein.
`
`The contributor grants a free, irrevocable license to the IEEE to incorporate text contained in this
`contribution, and any modifications thereof, in the creation of an IEEE Standards publication; to copyright
`in the IEEE’s name any IEEE Standards publication even though it may include portions of this
`contribution; and at the IEEE’s sole discretion to permit others to reproduce in whole or in part the
`resulting IEEE Standards publication. The contributor also acknowledges and accepts that this contribution
`may be made public by IEEE 802.16.
`
`The contributor is familiar with the IEEE 802.16 Patent Policy and Procedures (Version 1.0)
`<http://ieee802.org/16/ipr/patents/policy.html>, including the statement IEEE standards may include the
`known use of patent(s), including patent applications, if there is technical justification in the opinion of the
`standards-developing committee and provided the IEEE receives assurance from the patent holder that it
`will license applicants under reasonable terms and conditions for the purpose of implementing the
`standard.
`
`Early disclosure to the Working Group of patent information that might be relevant to the standard is
`essential to reduce the possibility for delays in the development process and increase the likelihood that the
`draft publication will be approved for publication. Please notify the Chair <mailto:r.b.marks@ieee.org> as
`early as possible, in written or electronic form, of any patents (granted or under application) that may cover
`technology that is under consideration by or has been approved by IEEE 802.16. The Chair will disclose
`this notification via the IEEE 802.16 web site
`<http://ieee802.org/16/ipr/patents/notices>.
`
`1
`
`HONDA 1017
`
`

`

`2001-11-12
`
`80216abc-01/61
`
` Fast and Efficient Bandwidth Request Mechanism for
`IEEE 802.16a OFDM PHY
`
`Jerry Krinock, Manoneet Singh, Mike Paff, Vincent Tien, Arvind Lonkar and Lawrence Fung
`Radia Communications
`
`1. MOTIVATION FOR THIS STUDY
`
`The IEEE 80216ab-01_01r2 working document states that access to the uplink air interface in the OFDM mode
`of 802.16a shall be contention based [1, pp. 196]; however, details of the contention mechanism to be used for
`this purpose have not yet been specified. No performance results or MAC primitives to support any particular
`mechanism have been provided either.
`
` In this contribution, we perform a detailed, PHY-level evaluation of various contention-based techniques that
`may be used for bandwidth request in the OFDM PHY. The result of our investigation is the development of an
`optimal, contention-based BW request mechanism, which we propose and describe here.
`
`2. INTRODUCTION
`
`The key desirable features of a BW Request mechanism in any data network can be heuristically summarized
`as follows [2]:
`
`•
`
`•
`
`•
`
`•
`
`It should be contention-based in order to handle bursty data traffic with low overhead.
`
`It should function robustly under the constraints imposed by the physical channel (in terms of multipath
`spread, interference from other users, etc.).
`
`It should guarantee low delays in order to support delay-sensitive traffic.
`
`It should (preferably) be optimized using the features of the underlying PHY, so as to leverage the
`opportunities afforded by a particular type of PHY layer. At the same time, it must be simple enough to
`ensure that neither Base Station (BS) nor SS (Subscriber Station) are computationally overloaded.
`
`We shall now examine various BW request schemes for IEEE802.16a in light of the above observations.
`
`2.1 Background
`
`The Single Carrier mode in the 802.16a PHY uses a slotted-ALOHA-based BW request mechanism [1, pp. 58].
`According to this method, any SS that requires access to the uplink air interface must contend by transmitting
`a short packet during a reserved portion of the uplink frame. Collision-free transmissions are received as valid
`requests at the BS, which can then allocate specific slots to successful SSs to uplink during subsequent frames.
`The basic motivation for using such a scheme comes from DOCSIS, where a similar mechanism is used for
`controlling subscriber access among several simultaneous, bursty data terminals [3].
`
` While the slotted ALOHA based method offers reasonable performance in the Single carrier mode, the
`efficiency of this scheme is dramatically reduced when combined with multicarrier transmission at the PHY layer
`[4]. This is because, unlike DOCSIS, where the contention slots are all of relatively short duration ( minislots ),
`an OFDM (OFDMA) symbol is, by definition, much longer than a single carrier symbol. Furthermore, decoding a
`contention packet requires the BS to have an estimate of the SS s channel, which would require an additional
`preamble to be transmitted before each contention packet, adding further redundancy to the request process.
`Both these factors put together drastically reduce1 the efficiency of the ALOHA based scheme for
`OFDM/OFDMA.
`
`
`1 We quantify this loss more precisely in Section 5.
`
`2
`
`

`

`2001-11-12
`
`80216abc-01/61
`
` In a multi-carrier transmission, besides the time dimension, the frequency dimension is also available to
`design a BW request scheme. This observation has been used in OFDMA mode of the 802.16a PHY to design
`a BW request mechanism based on CDMA codes [5]. The proposed scheme seeks to accommodate more
`than one contending user per OFDMA symbol by providing a set of frequency domain codes that control uplink
`access via contention. We have already analyzed the PHY level robustness of this scheme in an earlier
`contribution [6], and identified several issues.
`
` The scheme that we propose in this document is based on (time) differential modulation of codes across
`disjoint frequency partitions. The use of differential transmission effectively eliminates the influence of the
`frequency selective channel, so that the use of specific codes across (frequency) subcarriers can then help to
`multiplex several contending SSs within a single symbol. In the remainder of this document, we describe this
`scheme in detail, and study its PHY-level performance over the various SUI channels [7]. We show that a three-
`dimensional grid of time, frequency and codes provides optimal performance2 as compared to other approaches
`for BW access.
`
`3. DESCRIPTION OF PROPOSED MECHANISM
`
`The proposed mechanism operates by setting up a time-frequency partition of the contention window within any
`OFDM uplink frame (cf. Fig. 1):
`
`• Along time, symbols are grouped into pairs ( contention slots ) across which differential keying takes
`place. For purposes of description, the first symbol in each contention slot is labeled 0 , and the
`following symbol in the pair is labeled 1 .
`
`• Along frequency, each symbol is divided into sets of subcarriers ( bandwidth request channels ) that
`carry the bits of a differentially keyed code. The division of the available 200 subcarriers (for 256 point
`OFDM) into bandwidth request channels is a static one i.e. determined and specified once during
`system design3. For purposes of description, the total number of subcarriers in a bandwidth request
`channel is labeled K , and the index k is used to refer to a particular subcarrier in the set.
`
`To contend for bandwidth, a SS must choose, at random:
`
`(i)
`(ii)
`(iii)
`
`A contention slot within any particular uplink frame;
`A bandwidth request channel (i.e., a set of subcarriers) within that contention slot; and
`A K-bit code, to be transmitted differentially across the subcarriers in the given contention slot.
`
`For example, let b0(k,n) denote the first (randomly chosen) bit transmitted on the kth subcarrier in the nth
`contention slot. Then
`
`
`
` ),( nkb
`1
`
`=
`
`
`
` )( ),( nkbkC
`0
`m
`
`
`
`(1)
`
`denotes the bit transmitted on the same (kth) subcarrier during the second symbol in that contention slot. Here
`{Cm(k)}k are the bits of the (K-bit) mth code, selected by the SS to transmit in that frame.
`
`The transmitted bits b0(k,n) and b1(k,n) are then received by the BS (across the frequency-selective channel) as
`r0(k,n) and r1(k,n), where
`
`
`
` ),( nkr
`0
`
`
`
` ),( nkr
`1
`
`=
`
`=
`
`
`
` ),( ),( nkHnkb
`0
`
`
`
`
`
` ),( ),( nkHnkb
`1
`
`
`
`η+
`(
`
`k
`
`),
`
`η′+
`(
`
`k
`
`).
`
`(2)
`
`(3)
`
`)(kη′
`)(kη and
` denote uncorrelated additive Gaussian noise, and H(k,n) denotes the channel transfer
`Here
`function at the kth tone, assumed to be invariant across two OFDM symbols, i.e.,
`
`2 High efficiency, simple implementation, and low false alarm rate.
`3 We discuss some performance results for various divisions later on in this document.
`
`3
`
`

`

`1
`l
`τ and complex gain
`where we have considered an L path channel with delay
`nl ,
`associated with the l th path during the nth contention slot interval [7].
`
`80216abc-01/61
`
`(4)
`
`h
`,
`nl
`
`=
`
`I
`
`h
`,
`nl
`
` hj.+
`
`
`,
`nl
`
`Q
`
`=
`
`f
`
` Tk /
`
`
`s
`
`−
`
`τπ
`.
`.2
`f
`,
`nl
`
`j
`
`nl eh ,
`
`
`
`=∑
`
`L
`
`2001-11-12
`
`
`
` nkH ),(
`
`=
`
`It is now straightforward to detect the presence (or absence) of the code C(m) on any particular bandwidth-
`request channel. A simple detection strategy that may be used for this purpose is described in the Section 4.
`
`
`
`3.1 Notation
`
`Fig. 1: Contention Slots and BW Request Channels
`
`A formal definition of the various entities introduced in this section is as follows:
`
`We define a bandwidth-request signal as the two consecutive differentially encoded OFDM symbols required
`to attempt a bandwidth request.
`
`We define a contention slot as a time duration equal to that of two OFDM symbols. We do this because there
`are two consecutive OFDM symbols required for a contending user to transmit its bandwidth-request signal in
`the proposed Multichannel Bandwidth Request Scheme. (It is coincidental that a slotted ALOHA scheme would
`also require two consecutive OFDM symbols for contention; however, in that case, the first is a preamble for
`training, and the second is a MAC header.)
`
`We define a bandwidth-request code as sequence of bits from the alphabet {1, -1} which the SS selects at
`random from an available pool to use in coding its signal. In our proposed realization, this pool contains 32
`sixteen-bit codes. The first sixteen codes are the rows of a 16x16 Hadamard-Walsh matrix. The other sixteen
`codes are their bit-wise complements.
`
`4
`
`

`

`2001-11-12
`
`80216abc-01/61
`
`We define a bandwidth-request seed as another sequence of bits from the alphabet {1, -1} which the SS
`selects at random from another available pool. In our proposed realization4, this pool contains 256 sixteen-bit
`codes, which are generated from a pseudonoise generator by taking subsequent output sequences of 16 bits.
`
`We define a bandwidth-request channel as the set of tones that form one member of a partitioning of the
`available used OFDM tones in the uplink channel. In one proposed realization, we partition the available 200
`tones of the default 256-point OFDM uplink into twelve bandwidth-request channels of sixteen tones each. The
`tones in each bandwidth-request channel are distributed uniformly across the OFDM symbol, except that eight
`tones are unused5, since 16 x 12 = 192. Note that the number of carriers in each bandwidth-request channel
`must equal the number of bits in each bandwidth-request code and seed.
`
`4. DESCRIPTION OF THE SIMULATION
`
`Simulations were done to show the performance of the proposed Multichannel Bandwidth-Request Scheme at
`the PHY level.
`
`4.1 Stimulus Generation
`
`The simulation operates by first realizing a schedule of transmissions for the desired number of contention slots
`in the simulation. The simulation operates on a system sample time = (OFDM symbol duration) / (number of FFT
`points). At each sample time, a single bandwidth-request signal may or may not occur, depending on the
`probability of such a signal at the data point being simulated. Because each contention slot is two OFDM
`symbols, this probability is equal to the average number of contending transmissions per contention slot, divided
`by twice the number of FFT points. Note that allowing only one signal per sample ignores the possibility that
`there may be multiple signals at one sample time. However, since there are hundreds of samples and only a
`handful of average bandwidth-request signals in each contention slot, this assumption results in negligible error.
`
`For each bandwidth-request signal so generated, a bandwidth-request seed, a bandwidth-request code and a
`bandwidth-request channel are randomly selected from the available pools. Finally, a SS transmitter power
`level is randomly selected from the range of ± 1.0 dB from nominal, and the signal is marked to occur in the
`contention slot nearest the instant sample time. This latter behavior mimics actual SS operation, assuming that
`the SS are already ranged well enough to delay its bandwidth-request signal so that will be received within the
`cyclic-prefix tolerance of the system, and also that it knows which OFDM symbols are available for transmitting
`the first and second symbols of a bandwidth-request signal.
`
`After the schedule of transmissions is generated, the actual simulation is run, one contention slot at a time.
`
`The first step in the actual simulation is generating the received signals. For each contention slot, the active
`bandwidth-request signals are picked out of the schedule.
`
`For each active bandwidth-request signal, a ± 1 representing the BPSK modulation of each bit in the seed is
`generated for each tone of the its bandwidth-request channel. This begins the signal received in the first OFDM
`symbol. Similarly, a ± 1 representing the BPSK modulation of each bit in the seed is generated, but in this case
`the code bit is XORed with the corresponding seed bit, effecting differential modulation. This begins the signal
`received in the second OFDM symbol. All signals are multiplied by the randomly-selected tx power level.
`Finally, a random channel is realized from the desired SUI model, and the modulation in each tone, for both the
`first and second signals, is multiplied by the complex frequency-domain channel gain from the random channel
`realization. Thus the received signal due to each active bandwidth-request signal is two vectors of complex
`numbers, one for the first OFDM symbol and one for the second.
`
`If there is more than one active bandwidth-request signal in the instant contention slot, the received signals due
`to the others are generated in the same way, and all signals are added together to form a composite receive
`signal. Note that if all active bandwidth-request signals in a given contention slot have chosen different
`bandwidth-request channels, the additions will occur on different tones, and assuming no intercarrier
`interference in the BS receiver, there will therefore be no interference among these bandwidth-request signals.
`
`4 The choice of the seed bits {b0} does not seem to affect the performance of the proposed scheme.
`5 The eight unused tones can be used very effectively for initial ranging, as will be shown in the sequel.
`
`5
`
`

`

`2001-11-12
`
`4.2 Signal Detection
`
`80216abc-01/61
`
`After so generating the received signal for a given contention slot, a BS receiver is simulated for each available
`bandwidth-request channel, for each available code in in the contention slot. The receiver computes two
`metrics. The first metric measures the error in the received signal correlated to the expected code:
`
`errCode
`
`(
`
`m
`
`)
`
`= ∑
`1
`K
`
`k
`
`where
`
`−
` ),( )( ,),( nkrkCnkr
`0
`1
`m
`
`
`
`
`
`
`
`(5)
`
`K = number of carriers in a bandwidth- request channel (proposed = 16)
`k = index on carriers
`) =
`(
`n k
` BS rx FFT output, first OFDM symbol, contention slot n, carrier k
`r
`(
`) =
` BS rx FFT output, second OFDM symbol, contention slot n, carrier k
`n k
`r
`( ) =
` code bit, code m, bit k
`C km
`
`0 1
`
`, ,
`
`The second metric measures the error in the power level relative to the expected power level of 1.0:
`
`errPower
`
`(
`
`
`
`)m
`
`= ∑
`1
`K
`
`k
`
`
`
` ),( nkr
`0
`
`2
`
`−
`
`.1
`
`(6)
`
`) could have been used as well, since both signals suffer the same
`,(Note that, in the above expression, r k n
`2
`
`0
`tx power variation and same channel instance.
`
`Finally the two errors are added together to develop a sufficient statistic:
`
`errComposi
`
`
`
`(mte
`
`)
`
`=
`
`errCode
`
`(
`
`m
`
`)
`
`α+
`
`errPower
`
`(
`
`
`
`),m
`
` (7)
`
`which is then compared with a decision threshold to determine whether or not a bandwidth-request signal is
`declared to be detected or not.
`
`For each such receiver (contention slot, bandwidth-request channel, bandwidth-request code), if the
`errComposite exceeds the allowed threshold, the result is ignored. If it does not exceed the allowed threshold,
`the result is tallied as either a (valid) detection or a false alarm. To determine which, the schedule of active
`bandwidth-request transmissions is examined. If there was only one bandwidth-request signal transmitted
`during the instant contention slot, on the instant bandwidth-request channel, using the instant bandwidth-request
`code, a valid detection is tallied. Otherwise, a false alarm is tallied. Note that if more than one SS transmit a
`bandwidth-request signal in the same contention slot, on the same bandwidth-request channel, using the same
`bandwidth-request code, and the resulting receiver errComposite is within the allowed threshold, the BS will
`make an allocation which will be claimed by all SS sending such a signal, resulting in a subsequent collision.
`Our simulation correctly models such occurrences as false alarms.
`
`After all receivers in all contention slots have been thus simulated, the final tally of detections is divided by the
`total number of bandwidth-request signals transmitted in the schedule, resulting in the probability of detection.
`The final tally of false alarms is divided by the number of contention slots simulated, resulting in the number of
`false alarms per contention slot.
`
`4.3 Summary of Basic Parameters
`
`The basic parameters common to all simulation results in this study are shown in Table 1.
`
`6
`
`

`

`2001-11-12
`
`80216abc-01/61
`
`Number of Points In FFT
`Number of lower-frequency guard tones
`Number of higher-frequency guard tones
`Channel Width
`Subcarrier Frequency Spacing
`
`Ratio of signal power to additive white Gaussian noise in
`each receiver FFT output bin
`Distribution ot SS received power levels
`Channel Model
`Number of Contention Slots simulated per scenario
`
` Table 1. Basic Parameters Common To All Simulations
`
`256
`28
`27
`6.0 MHz
`(7/6)*(6.0MHz/256) =
`27.34 KHz
`Varied from 15-99 dB
`
`Uniform over ±1.0 dB
`SUI-1, SUI-4, SUI-6
`1000 to 5000
`
`4.4 Modeling of Slotted ALOHA for Comparison
`
`The proposed scheme is similar to conventional slotted ALOHA when the number of request channels is
`reduced to one. Therefore, our simulation engine can also be used to make a fair comparison with slotted
`ALOHA. The only difference in the model is that while our model sends two differentially coded signals in the
`two symbols of the contention slot, a slotted ALOHA system sends a preamble and MAC header instead.
`However, it was verified, not surprisingly, that the 200-bit codes in our model perform exactly as the preamble
`and MAC header; that is, signals are always detected when there is no collision, and never detected when there
`is a collision. Thus, the slotted ALOHA system can be modeled, for all practical purposes, as the proposed
`Multichannel bandwidth-request system with all used carriers allocated to a single bandwidth-request channel.
`The parameters used for each system are shown in Table 2.
`
`Walsh codes are the rows of a square Hadamard-Walsh matrix. Complemented-Walsh codes are their bit-wise
`complements.
`
`Bandwidth Request
`Scheme
`Number of carriers
`per request channel
`Number of
`bandwidth-request
`channels (for 256 pt.
`OFDM)
`Codes
`
`Conventional slotted
`ALOHA
`200
`
`1
`
`8
`
`24
`
`Proposed Scheme - Options
`
`16
`
`12
`
`32
`
`6
`
`One 200-bit PN codes
`(used for convenience;
`the code choice has
`no effect)
`
`16 8-bit Walsh
`and
`complemented-
`Walsh codes
`
`32 16-bit Walsh
`and
`complemented-
`Walsh codes
`
`64 32--bit Walsh
`and
`complemented-
`Walsh codes
`
` Table 2. Simulation Parameters For Proposed Scheme vs. Slotted ALOHA
`
`4.5 Retransmissions
`
`To verify the results, we extended the simulation to include retransmissions. For these simulations, we set the
`threshold to the optimum value previously determined, and then triggered an additional bandwidth-request
`signal to be transmitted whenever one failed to be detected. A simple backoff algorithm was used, with the
`retransmission equally likely to occur in either of the following three contention slots. Again, the bandwidth-
`request code and bandwidth-request channel were selected at random from the available pools.
`
`7
`
`

`

`2001-11-12
`
`80216abc-01/61
`
`5. SIMULATION RESULTS
`
`5.1 Raw Performance (no retransmissions)
`
`The raw detection probability is a first measure of contention system performance. In this case, retransmissions
`are disabled, and thus the receiver sees only the constant average stream of bandwidth-request signals from
`the originally generated schedule.
`
`We begin with the results obtained using 32-bit bandwidth-request codes on 6 bandwidth-request channels.
`Figure 2 shows the performance at high SNR. The green curves beginning at the top left are the detection
`probability indexed on the left axis, as is the blue slotted-ALOHA curve. The red curves are the false alarm
`rates indexed on the right axis. Note that the plot has been stopped where the probability of detection falls to
`75%, since the system is becoming unuseable at this point, and the BS must start restricting traffic. Using this
`criteria, it can be seen that the slotted-ALOHA system can handle only a fifth of the traffic carried by the
`proposed scheme. Equivalently, the BS must schedule five times as many contention slots to achieve the same
`throughput as the proposed scheme.
`
`Figure 2. Performance, Proposed System with 32-bit codes, 99 dB SNR, no retransmits.
`
`The false alarm rates in this case were used to set the receive decoder s threshold, as explained in Section 6.3,
`for this code length.
`
`Figure 3 shows the performance with 15 dB additive white Gaussian noise (AWGN). Note that both detections
`and false alarms are reduced in this very noisy environment, with the false alarms in particular falling to zero.
`Obviously, if the threshold could be adapted as a function of SNR, our proposed scheme would perform even
`better. However, that may be difficult to do, since we are decoding short bursts, only two OFDM symbols, of
`unknown codes from unknown users.
`
`8
`
`

`

`2001-11-12
`
`80216abc-01/61
`
`Figure 3. Performance, Proposed System with 32-bit codes, 15 dB SNR, no retransmits.
`
`It is assumed in all figures that the slotted-ALOHA system will be completely unaffected by channel response
`and noise. That is why the blue curve and purple baseline showing, respectively, detections and false alarms,
`are the same in all figures. The blue curve agrees with some closed-form back-of-an-envelope analysis we
`have performed for this simple case, helping to validate our simulation engine.
`
`Figs. 4 through 6 show the performance of the system using 16-bit codes. Note that in these cases the
`abscissa is extended to 4 average bandwidth-request signals per contention slot, to show the additional
`capacity. This is attributed to doubling the available number of bandwidth-request channels from six to twelve.
`
`Finally, Figure 7 and Figure 8 show performance with 8-bit codes. Note the further extension of the abscissa.
`Figure 7 is stunning, showing how the proposed system can sift out six users per contention slot with 75%
`probability, while conventional slotted-ALOHA has completely died. This performance comes with a noise
`penalty, however, which is to be expected.
`
`9
`
`

`

`2001-11-12
`
`80216abc-01/61
`
`Figure 4. Performance, Proposed System with 16-bit codes, 99 dB SNR, no retransmits.
`
`Figure 5. Performance, Proposed System with 16-bit codes, 21 dB SNR, no retransmits.
`
`10
`
`

`

`2001-11-12
`
`80216abc-01/61
`
`Figure 6. Performance, Proposed System with 16-bit codes, 15 dB SNR, no retransmits.
`
`Figure 7. Performance, Proposed System with 8-bit codes, 99 dB SNR, no retransmits.
`
`11
`
`

`

`2001-11-12
`
`80216abc-01/61
`
`Figure 8. Performance, Proposed System with 8-bit codes, 27 dB SNR, no retransmits.
`
`5.2 System Performance with Retransmissions
`
`In actual system operation, the BS must control congestion in the contention channels by adjusting the
`retransmission backoff parameters and frequency of contention slots. As congestion increases, the BS may
`allocate more symbols for contention, until the UL capacity is fully utilized, but then must increase the minimum
`and maximum backoff exponents, limiting traffic.
`
`To get some idea how the different schemes would perform in practice, simulations were run with
`retransmissions enabled as described in Section 4.5. The fixed backoff exponents do not allow the BS to
`dynamically limit traffic, and although such a system would eventually melt down in practice, it was used to
`compare performance of the two schemes.
`
`Simulations were configured for each of the schemes. The proposed scheme was operated with 16-bit codes
`over a SUI-4 channel. To measure when the system was starting to melt down , triggers were set to tally a
`counter each time that more than twenty bandwidth-request signals were transmitted simultaneously on a single
`bandwidth-request channel. Twenty transmissions contending for a channel which can only carry nominally one
`is a ridiculous number, and this is taken as an indication that indeed things are starting to melt down . After
`512 such counts, a melt down was declared and the simulation aborted.
`
`To compare the two schemes, simulations of 2,500 contention slots duration were run successively with each
`bandwidth-request scheme. The melt-down counter was reset at the beginning of each simulation. In each
`simulation, the average number of contending bandwidth-request signals per contention slot (the abscissa of all
`figures in the previous section) was increased from the prior simulation. In the first simulation, it was 0.1. In the
`eight subsequent simulations, it was 0.2, 0.4, 0.8, 1.6, 2.0, 3.0, 4.0, 5.0 and 6.0 signals per contention slot.
`
`The result was that the conventional slotted ALOHA scheme melted down with 0.2 signals per slot, while the
`proposed scheme did not melt down until 2.0 signals per slot with 15 dB SNR, and not until 6.0 signals per slot
`with 99 dB SNR.
`
`12
`
`

`

`2001-11-12
`
`6. DISCUSSION
`
`6.1 Multipath Robustness
`
`80216abc-01/61
`
`We have shown in an earlier work [6] that the performance of code-based schemes on dispersive channels can
`suffer dramatically due to the frequency selectivity introduced by the channel. That study did not consider any
`additive noise, and assumed perfect power control. In the presence of such impairments, the performance can
`be expected to be further degraded.
`
`However, it is easy to see that the proposed scheme is not affected by the presence of a frequency selective
`channel, as seen in the similarity of the curves for SUI-1 and SUI-4. The performance in SUI-6 is worse but this
`is due to the fact that the present standard [1] does not accommodate the delay spread encountered on this
`channel even for its largest Cyclic prefix size. The use of differential modulation is able to show acceptable
`performance as far as bandwidth requests are concerned, even on this channel.
`
`6.2 Optimum Code Length
`
`It is apparent from the discussion and results in sec. 5 that code lengths of 8, 16 or 32 bits offer a steady
`tradeoff between capacity and robustness. In all cases, the performance is superior (by 5-20 times), when
`compared to the conventional method.
`
`6.3 False Alarm Pathology
`
`In our simulations, we tally false alarms into three categories. Recall that the BS receiver is, in fact, composed
`of multiple receivers , in a matrix of bandwidth-request channels times bandwidth-request codes. Three types
`of false alarms may be identified:
`
`(a) False alarms which occur when no bandwidth-request signal as transmitted on the receiving bandwidth-
`request channel. These false alarms are caused by noise only.
`
`(b) False alarms which occur when bandwidth-request signal(s) were transmitted on the receiving
`bandwidth-request channel, but none of them were of the receiving code.
`
`(c) False alarms which occur when two or more stations actually did transmit on the receiving bandwidth-
`request channel, using the receiving bandwidth-request code. However, these are tallied as false
`alarms since they are useless to the operator, and in fact the most undesirable type of false alarm. they
`will cause the BS to send a ranging response (RNG-RSP) message which will be answered by multiple
`SS in a subsequent collision.
`
`A typical breakdown of false alarms is shown in Figure 9.
`
`For completeness, note that between cases (b) and (c) are the cases of true detections, when only one
`bandwidth-request signal was transmitted using the receiving bandwidth-request code on the receiving
`bandwidth-request channel. If a signal such as this is detected, it is a true detection, whether or not there were
`other codes interfering with it. In most cases, interference from these other codes will prevent detection.
`
`13
`
`

`

`2001-11-12
`
`80216abc-01/61
`
`Figure 9. Typical Pathology of False Alarms. SUI-4 Channel, 15 dB SNR.
`
`6.4 Receiver Threshold Setting
`
`In simulating a receiver as we have described, the allowed threshold of errComposite setting is, of course,
`critical. Higher thresholds increase detection probabilities but also increase false alarms; lower threshold
`settings do the opposite. We assume that a conventional slotted-ALOHA system will have no false alarms; that
`is, it is virtually impossible for an OFDM preamble and MAC header to combine with other interfering
`preamble(s) and MAC header(s) and/or noise to be detected as a valid preamble and MAC header by the BS.
`The essence of our proposed scheme, however, is use the contention resource as efficiently as possible instead
`of using overkill. In such an efficient system, false alarms are possible. We expected from experience, and
`verified, that false alarms are increased at high signal-to-noise (S/N) ratio, so we set our thresholds for each
`code-length option by requiring that the false alarm rate be acceptably low when operating at a high SNR of 99
`dB. We have somewhat arbitrarily set our criteria to the following: False alarms shall occur at a rate less than
`.01 per contention slot, with the traffic frequency adjusted so that the ultimate raw probability of detection, i.e.
`that limited by collisions and not threshold setting, is greater than 75%. The latter condition comes from the
`realization that once traffic has increased to the point where the raw detection probability is less than 75%, the
`system is definitely in the danger zone and the BS is going to have to act fast to reduce congestion. The former
`condition of .01 false alarms per contention slot is admittedly, arbitrary. It seems that such a number would not
`impose too much congestion and is thought to be conservative.
`
`It was found that the longer the code (recall that we simulated 32, 16 and 8 bits), the higher the error tolerance
`threshold should be.
`
`We also tried some simulations using fewer codes, and using more codes which are available from a
`pseudonoise (PN) generator. The performance was found to be not as good, and these studies were
`discontinued.
`
`14
`
`

`

`2001-11-12
`
`7. CONCLUSIONS AND REMARKS
`
`80216abc-01/61
`
`A fast and efficient bandwidth request mechanism is presented for the OFDM mode of the IEEE