Co-existence of dissimilar wireless systems page 1 Co-existence of Dissimilar Wireless Systems Jan Kruys, Cisco Systems Third draft: December 3, 2003 This contribution has been prepared to assist the work of the Wi-Fi Alliance. It reflects the views of the author and does not contain or represent a policy or a commitment of Cisco Systems. Abstract There are two dimensions to spectrum sharing: vertical sharing: between systems with different levels of regulatory status (e.g. newcomers that have to live together with incumbents) and horizontal sharing: between systems with equal regulatory status. This document introduces the concept of vertical sharing but addresses horizontal sharing in some detail. A regulatory regime for vertical sharing of necessity has to be specific in that it deals with protection criteria and their application for a given service or system, with the intent to protect it from interference by other systems. A regulatory regime for the horizontal co-existence of diverse RF systems in the same, license exempt frequency space, is a tough nut to crack and so far it has eluded a satisfactory solution. However, a solution becomes more urgent as the FCC and other regulators plan to make more spectrum “commons” available under license exempt or light licensing rules. This short paper attempts to address the problem by reducing its scope to systems that access the medium in a non-deterministic manner. Noting that the main criterion is the receiver performance in the face of interference, it proposes a set of simple parameters and rules that could be the basis for regulatory measures aimed at “commons spectrum”. The rules are based on the rewarding efficient spectrum use by good receiver design and/or short medium occupation. The reward is given by a modification of the listen before talk criteria and the power output criteria. Thus, systems designers have choice: limit the time on the medium or improve the receiver, or both. Improving the receiver is understood to include the choice of modulation scheme as well as other aspects that affect the transmitter. The required behavior, is, of necessity, statistical in nature. Introduction Wireless telecommunications and wireless connectivity have seen a very high adaptation over the last decade and have proven to be significant contributors to economic growth. Two types of systems have dominated the scene: the mobile phone and the wireless LAN. The former relies on licensed – that is “exclusive use” – spectrum whereas the latter operates and indeed owns its success to “common use” of the same spectrum by all users. The merits and drawbacks have been argued and analysed in many ways elsewhere (see [1] for an example) and therefore that matter is not addressed here. Suffice it to state that both sorts of usage have their proper place in society. This paper focuses on the commons problem. There are two dimensions to spectrum sharing: vertical sharing: between systems with different levels of regulatory status (e.g. newcomers that have to live together with incumbents) and horizontal sharing: between systems with equal regulatory status. This document recognizes the problem of vertical sharing but addresses only horizontal sharing in a commons setting. The FCC and spectrum regulators outside the US are about to open up more spectrum for a “commons” type use under license exempt or light licensing rules. This short paper attempts to address the problem by reducing its scope to systems that access the medium in a nondeterministic manner. Noting that the main criterion is the receiver performance in the face of interference, it proposes a set of simple parameters and rules that could be the basis for Copyright © Cisco Systems Inc. Co-existence of dissimilar wireless systems page 2 regulatory measures aimed at “commons spectrum”. Such rules and the behaviour they assume a refine are of necessity statistical in nature. Deterministic behaviour in a shared environment is only possible when all devices implement the same set of rules and operate on the same time base. The obvious consequence is stifling of evolution and innovation. Vertical Sharing A regulatory regime for vertical sharing of necessity has to be specific in that it deals with the protection criteria and their application to an existing or a new service or application with the objective of protecting it from interference by other systems. A case in point is radar systems in the 5 GHz which are protected from interference by wireless devices by a detection and avoidance mechanism (DFS) [2] built into the latter. Other examples include the detection of broadcast transmissions by short range devices so as to allow them to operate in spectrum that is locally not in use by the broadcast services or that is locally useless, e.g. because of shadowing of an area by a the horizon or a range of hills. In both cases, the sharing takes place between very dissimilar systems with vastly differing operating range. In both cases sensitive receivers must be protected but in one case these receivers are few in number and co-located with powerful transmitters, in the other case the receivers are distributed and numerous. The detection capabilities of the “newcomer” devices must be tuned to the specific transmissions of the incumbents and therefore they will be different from case to case. Although this is a rich field that is just beginning to be investigated, this subject is not further addressed in this paper. Horizontal Sharing 1 Problem Summary Unlicensed spectrum, more recently more fashionably known as spectrum commons, pose an interesting problem: how to efficiently share spectrum without imposing technically unfair and/or stifling regulation. The following concentrates on the issue of sharing a “channel”. Although avoiding cochannel operation is high useful and in fact mandatory in some cases [2], doing so in a commons spectrum context is at once obvious (look for a clean piece of spectrum) and complex – in the commons spectrum, channels do not exist and any system can use any piece of spectrum. In fact, further work may well show that avoidance of co-channel operation in a commons world is a trivial and a reasonable on its own only if the commons a large – or sparsely used. The sharing study done by Microsoft [3] sums up the problem: interference varies with power levels, bandwidth ratios, and medium access methods and without any means of policing how systems behave, wide bandwidth devices with listen before talk medium access procedures, typically hold the short end of the stick. This means that applications that require significant wireless bandwidth are subject to the threat of punishing interference that turns the users experience into something not worth repeating. The broader view is that different applications require different bandwidth – or at least different signalling. It will be clear that some constraints are required on the type of systems that are to co-exist in the same spectrum: operating range has to be of the same order of magnitude and two way communication is required. Copyright © Cisco Systems Inc. Co-existence of dissimilar wireless systems page 3 In the absence of deterministic sharing mechanisms used in the “exclusive use” domain, the problem is finding means to encourage efficient use of spectrum without regulating specific parameters, or at least minimizing the number of regulated parameters. This is known as the problem of “the tragedy of the commons”. In the commons of old, the tragedy was overgrazing of the available shared pasture. In the spectrum domain today, the tragedy is too much interference – made worse by a wide disparity between the systems using the spectrum. In the commons of old, social control made sure that the tragedy was avoided. No such tools are available to the wireless systems designer or user. There are two categorically different approaches to solving the commons sharing problem: the heads up approach and the heads down approach. The heads-up approach requires a separate channel on which stations post their actual and intended use of a given piece of spectrum and all adjust their usage accordingly. The problem here is how to handle individual reductions to allow large numbers of users. This problem is not addressed here. Instead, we focus on the heads down approach: all stations look only after their own interests – but they play by the rules. Finally, it should be noted that there are two types of commons – single species commons and multi-species commons. Whereas the former allows sophisticated and efficient solutions - see IEEE802.11 and offspring – the latter, the broader common problem, is less tractable and cannot be mastered with the same efficient results as its narrower variant. Thus the rules developed here are rules of last resort, needed to address the hard problem of a heads-down, multiple species commons, suitable for the ISM and the U-NII band. 2 Analysis of the broader spectrum commons problem Interference is not a transmitter issue and it is not solved by listen before talk rules. Instead, it is a receiver issue. If receivers are able to distil the wanted signal from strong interference, there would be no problem. In fact, there would be no interference. The real problem is that receivers need a positive operating margin (or signal to noise + interference ratio). The higher the signalling rate, the higher the receiver’s operating margin has to be. The available margin, in turn, is determined by the strength of the received signal and the relative strength of the interference at the receiver. Note: even in small scale systems like RLANs the above is true – it is the reason for the RTS/CTS protocol of IEEE802.11. The better receiver is able to correctly receive weaker signals in the presence of more interference. However, better receivers can be costly (MIMO leads to higher power consumption, beam forming leads to higher device cost and operational limitations). To avoid that cost, the receiver can move closer to the sender and achieve the same result. The level of interference is determined by the power output of transmitters. That power in turn is driven by the need to deliver a certain power level at the intended receiver. Typically, the transmitter power is fixed by regulatory means or by convention (as in the case of Wi-Fi devices). Since the conventional wisdom is “more is better”, transmitter power is driven up by considerations of range and data rate. In fact, higher date rates are more energy efficient for both sender and receiver. A prisoner dilemma results: reducing power voluntarily will not benefit the volunteer so why should he do so. Copyright © Cisco Systems Inc. Co-existence of dissimilar wireless systems page 4 If the transmitter has no inherent incentive to reduce power and the receiver has no inherent incentive to improve its operating margin, one cannot rely on self interest to solve the commons problem and the only recourse is regulation. 3 What sort of rules are feasible? The first question is “what sort of rules are possible?” The nature of the problem is not so much complexity but lack of complete information at any point in time. Since the coexisting systems are assumed to be diverse, exact behaviour like in IEEE 802.11 can not be specified. In addition, the medium is not reliable and its properties as seen by any device, can change rapidly. Therefore a deterministic approach is not possible and the rules have to be aim at controlling the statistics of device behaviour and they have to be make use of input parameter values who values are (short term) averages – like duty cycles or transmission error rates. That statistical approach should be kept in mind in the following. The integration period for these statistics has to be long compared to the events being controlled. 4 What to regulate – what is a minimum set of rules for the spectrum commons? Which parameters should be regulated? Transmitter power and time patterns are major factors in creating interference and those have to be used as the targets of the regulatory constraints. Ways must be found to limit both with undue restrictions on technical solutions. Systems designers must be encouraged to trade time in the air for power output and vice versa – in other words the power limit must contain a duty cycle factor, e.g. TXout*Duty Cycle = constant. Systems that can do with a lower TXout can be in the air longer and systems that are in the air for a short time can use a higher TXout. The fact that broadband systems suffer more from narrowband interference than vice versa suggests that the appropriate power limit is power spectral density, not power as such. The spatial aspect also should not be ignored. Interference is caused over a certain area and that area is largely independent of the antenna gain – if one ignores the vertical dimension. Therefore the proper power measure is PSD at the transmitter output. The next step is put constraints on when a transmitter may be turned on. Although the PSD/duty cycle limit motivates designers to be frugal with power, it does not help to separate transmissions in time. Therefore a Listen-Before-Talk requirement is useful. The main rule is that if the spectrum to be used is sensed to have energy in it, the transmitter will wait a random time period before trying again, e.g. n*2 usec where n = a random variable. This variable too can be made flexible: at low duty cycles, a fixed backoff can be considered sufficient, at high duty cycles, a longer back-off would prevent high duty cycle systems from locking out low duty cycle systems. Since we have to make rules covering a very wide range of systems, it is hard to set a single limit for all. Again, the usage patterns can be used to increase inter-system fairness (= to increase the freedom of designers). However pure LBT is inefficient at high utilization levels and if we want to avoid getting into specification of medium access rules, the LBT principle has to be modified slightly: if a transmitter expects that a transmission will be effective, it need not wait for a free channel. This is achieved by a modification of the LBT threshold (higher success expectation should allow the threshold to be set higher = less constrained access). Copyright © Cisco Systems Inc. Co-existence of dissimilar wireless systems page 5 Finally, for some applications it may be advantageous to lump a short sequence of transmissions as one LBT event. To make this possible without undue burden on other users, the duration of such a burst has to be limited. Considering the broad trend towards higher transmission rates, that burst period can be kept short. In summary we have three basic rules: a) All stations observe a common PSD limit - modified by their on airtime - less time in the air means more power b) All stations shall use LBT – modified by their PSD and receiver success rate - a lower PSD output allows a higher LBT threshold and higher probability of successful receive allows a being less sensitive to other transmissions. c) The maximum time a station can transmit or otherwise occupy the medium is limited In the following we look at parameter values for a simple set of Spectrum Commons rules and their consequences. Note that we only consider systems with equal status. National Security, safety of life and other considerations will affect the values of the regulated parameters. However, the principle can remain the same. 5 Example of Spectrum Commons Rules and Parameter values - Parameter description PSD limit: Power spectral density in mW – the average power while the transmitter is on, measured at the antenna connector. In operations, the PSD limit is modified by the duty cycle of the transmitter and the receive failure expectation to yield the actual PSD value for a given transmission. Duty cycle: The ratio between on-air time and the separation between onair events. The events considered are events in one direction between a pair of devices. Either device may communicate with other devices – the duty cycles of those events are considered separately. The on-air time used to determine the duty cycle covers all the time between gaining access to the spectrum after an LBT sensing until the spectrum is released. LBT threshold Listen before talk reduces unnecessary interference. However, LBT can lead to inefficiency – sometimes waiting for a quiet channel is not needed - e.g. if the receiver is nearby or has a low (=good) operating margin. Thus a high actual PSD would need a lower, more sensitive threshold and vice versa. LBT Back-off When the spectrum is occupied, the transmitter has to wait for a new opportunity. However, the wait period until that new attempt should reflect the failure expectation (lower means less back-off). Rx Failure Expectation Many systems are designed with an acceptable “receive event” error rate in mind. This value can be used as parameter to drive transmit decisions: if the expectation of failure is high, transmission should be discouraged. Transmitters can be informed by receivers of the RFE – Copyright © Cisco Systems Inc. Co-existence of dissimilar wireless systems page 6 instantaneous accuracy is less important - statistical accuracy is sufficient. - Regulated Parameters, values and modifiers PSD limit and LBT threshold are defined in terms of a reference case: 16Mb/s in a 20 MHz channel over 32 m in free space with a good, conventional receiver. The modifier definitions describe how the nominal values of the regulated parameters are converted into actual values, depending on the local conditions. Absolute Tx out limit Absolute air time limit Integration period 30 dBm (or less, as in the case on the lower two U-NII bands) 2 msec (one transmission or burst of transmissions not separated by LBT medium sensing) > 1 second Nominal PSD limit Nominal LBT threshold Nominal LBT back-off PSDnom = 0 dBm/MHz (-10 dBm/100KHz, etc) LBTnom = -80 dBm/MHz LBOnom = (r+s*n)usec, r < LBOnom < 15 s - is a scale factor that allows the rules to be adapted to various operating range conditions, e.g. for 100 m, s=1 usec. r is a constant that models device delay in sensing the medium – 5 usec is a suitable value. Nominal Duty Cycle Nominal RFE value DCnom = .1 RFEnom = .1 DC modifier RFE Modifier DCmod = (DCact/DCnom) RFEmod = (RFEact/RFEnom) Actual RFE is determined dynamically from receiver feedback which is system specific. At duty cycles below DCnom, RFEmod may be set to .1 Actual PSD value Actual LBT threshold Actual LBT back-off PSDact = PSDnom - 10logDCmod LBTact = LBTnom - 10logRFEmod - 10logDCmod + (PSDnom-PSDact) LBTact = (LBOnom*RFE mod)usec, r < LBTact < 63 The effect of these values and rules is that: - no device can monopolize the medium - shorter duty cycles facilitate access to the RF spectrum and/or higher RF power - better receivers give advantageous channel access. - channel hogging to get a message or packet through is discouraged. These rules do not prevent systems to use fast acknowledgment nor do they prevent a reasonable measure of QoS – schemes like WME of IEEE802.11 will work, as will bursting, high gain antennas and MIMO schemes. 6 Some test cases a) Telemetry system, 1 MHz, 1Mb/s, 2.4GHz, 10% duty cycle, 128 m range indoor: Available link budget: 0 dBm – 114 dBm noise floor = 114 dB Required link budget: pathloss + SNR +Implementation margin = 41+58+6+6 = 111 dB. Copyright © Cisco Systems Inc. Co-existence of dissimilar wireless systems page 7 b) Same system, duty cycle = 1% allows 10dBm duty cycle advantage, doubles the range to 256 m in an indoor environment or more in an outdoor environment. c) RFID system, 2,4 GHz duty cycle = 2%, range is 4 m, passive tag with signal gain of -50 dB (assumed!), Rx sensitivity = -114 + 12 dB = -102 dBm Txout = 0 dBm with a 7dBm DC advantage (2% instead of 10% duty cycle) Link budget: 7 – (-102) = 109 dBm At 4 m, two way pathloss = 65 dB; margin for tag is 44dB but it needs 50dB. This requires quadrupling the bandwidth to make it work at the targeted range. d) RLAN, Wi-Fi, 5GHz, 10 stations, 24 Mb/s, 200mW, average duty cycle per station is 2 %, SNR = 18dB, impl. margin is 8 dB. Max Tx out is 12dBm + 7dBm Duty Cycle advantage Required indoor range: 64m Available link budget: 19dBm – (101dBm noise floor) = 120dB. Required link budget: pathloss + SNR + Impl margin = 46+48+18+8 = 120 dB. e) Same RLAN, 2x2 MIMO, 10 dB Rx improvement Range increases to 128m or LBT threshold can be increased by 10dB to gain throughput advantage. 7 Implementation aspects The proposed co-existence rules above may seem complex and relatively costly to implement. However, when compared to low cost devices on the market today like Wi-Fi and Bluetooth, the above is seen to be simpler and relatively easy to implement. Designers can make choices that obtain the maximum performance possible or the can limit device performance and behavior so to reduce cost. For example, devices with a constant traffic rate need not calculate their own statistics and a fixed duty cycle means the Tx power can be fixed as well. Designers may also opt for higher performance, more costly solutions such as beamforming antennas. Such antennas allow higher data rates as well as a lower RFE – both factors allow them to decrease their LBT threshold and LBT back-off and thus get preferential spectrum access. Receiver performance can be ignored but in that case performance is limited by the nominal power and duty cycle values. 8 Conclusion The parameters and rules given in this paper offer the prospect of improving spectrum sharing among diverse systems. By linking power spectral density, duty cycle and medium access constraints, this approach forces designers to make choices that optimize the behavior of a system in a given application and cost profile but that are neutral from the co-existence point of view. A small set of “test cases” confirms that proposed regime does not constrain systems design or applications unduly. References: 1] Werbach Paper on Commons versus Exclusive rights spectrum rules 2] ITU-R Recommendation M1562 – Characteristics of DFS for the protection of the radio-determination service. 3] Microsoft …..paper on co-existence… Copyright © Cisco Systems Inc. Co-existence of dissimilar wireless systems page 8 Appendix: Determining RFE First of all it should be noted that the underlying assumption in all of the preceding is that it is not possible to regulate device behavior in a deterministic, case specific manner. Interference in common spectrum is statistical in nature and therefore the RFE parameter can never be an accurate predictor of the actual chance of communication failure. Thus the means of determining RFE need not be accurate and microsecond accurate either. This argument can be taken a step further: there is no need to specify a mechanism for determining RFE. Each system with have its preferred solution or method that give optimum performance for that system. Important from a regulatory viewpoint is that the mechanism is present and effective. Implementors have the option of setting RFE at .1 in case they want to avoid the expense of implementing a dynamic mechanism. The following gives some examples. 1) In-band signaling with each transmission This method adds a bit of overhead to each transmission but it has the advantage that is keeps a running tab on actual conditions. This method does not work well with very bursty traffic. 2) Simple acknowledgement. Keeping acknowledgement statistics will work in many cases where changes in the RF channel occur slowly relative to the channel occupancy. Many indoor environments have sufficient reflectivity to “smear out” RF fields so that the above condition is likely to occur. 3) Broadcasting local interference statistics. Where communications are intermittent, a feed forward mechanism may be employed: receivers can broadcast their typical or averaged interference levels so that transmitters have a basis for picking a reasonable RFE value for a given transmitter. Notably in the case of RLAN access points this may be a suitable mechanism - the interference data can be sent out with each beacon. Note that it is possible to verify claims about a dynamic RFE implementation: by setting up a communication between two devices and injecting interference at the receiver, the behavior of the transmitter can be made to change in predictable ways. References [1] Werbach, Open Spectrum: The New Wireless Paradigm. Spectrum Series Working Paper #6 October 2002. [2] ITU-R Recommendation M1562 – Characteristics of DFS for the protection of the radio-determination service. [3] Pierre De Vries & Amer Hassan, “Spectrum Sharing Rules for New Unlicensed Bands,” unpublished, Dec 2003. Copyright © Cisco Systems Inc.