P25 debate: The digital standard revisited
Editor’s note: MRT’s April 2001 article, “Does the Digital Radio Standard Come Up Short?” by Stephen Bartlett, drew strong response from the standards-creation community. A critique and technical assessment of the article has been prepared on behalf of P25 proponents by Bernie Olson, chairman of the Telecommunications Industry Association TR8.18 Committee on Compatibility and Interference. Following his analysis, we also present a further clarification by author Bartlett of his position.
The article “Does the Digital Radio Standard Come Up Short?” does a gross disservice to Project 25. Although there are some good technical descriptions of each modulation, the author’s comparison uses different performance criteria for the three modulations.
He then compounds the issue by using free-space propagation loss for a range comparison. Free-space loss is only valid for point-to-point paths where 0.6 of a Fresnel zone clearance is provided. In the land-mobile environment, this is commonly not the case. Therefore, the propagation power-loss exponent is closer to four than the free-space value of two, as used in the author’s comparison. This use of different performance criteria and an extremely liberal propagation loss model has resulted in a distorted conclusion that cannot be supported in an actual land-mobile radio environment.
The flawed methodology used by the author is further aggravated by the use of information from different sources. The author did not acknowledge any disparity among the various reference levels, specifications or derived values from the different sources. Tables 1 and 2 on page 20 clearly show the difference between the assumptions used by the author and those embodied in TIA-TSB88-A. The first table shows the relative coverage range differences based on the criteria values from “TIA/EIA Telecommunications Systems Bulletin TSB88-A, June 1999.” These values are based on a propagation power-loss exponent of 4, more typical of land-mobile communications. The derivation of these values is presented later. The second table is extracted from the author’s chart. The receiver IF bandwidth is indicated for the different scenarios.
The results in Table 1 are dramatically different from the author’s conclusion chart (Table 2).
The author bases his range arguments on basic carrier-to-noise criteria derived from various references. However, when the author calculates C/N from receiver bandwidth and published receiver threshold “spec” data, he fails to account for receiver noise figure. One could leave receiver noise figure out of the picture and reference all performance to the receiver (demodulator) input, as the author attempts to do. This can, and does, distort the results unless a common noise figure reference is used.
Two receiver sensitivities (static and faded) are given in TIA-102 publications. The static sensitivity is used to make laboratory or bench measurements. For analog radios, this benchmark is 12dB SINAD. For digital radios, this benchmark is 5% bit-error rate. Static sensitivity is not used in predicting coverage or range. It is used to ensure that the receivers have a published performance specification that customers can measure and use to compare receivers from different vendors. This is the receiver sensitivity value that normally appears on manufacturer’s specification sheets.
The TSB88-A, TIA102.CAAA and TIA102.CAAB documents also include faded sensitivity values. There is no analog faded sensitivity standard, but the median of the values used by various system designers is 13dB above the 12dB SINAD value for performance. For digital radios, the typical value to achieve 5% BER in faded conditions is 6dB to 8dB above the 5% BER static sensitivity level. However, 5% BER in a faded environment is insufficient to produce acceptable audio. A lower BER is required to achieve acceptable audio quality levels, with 2.6% BER being considered the minimum design level for Project 25 digital radios using the IMBE vocoder.
The author apparently used pieces of different specifications but failed to recognize the different scenarios these values are intended to represent. Also, many of the values cited for C/N are referenced relative to the receiver’s “thermal noise threshold,” which is the sum of the receiver’s noise figure and the thermal noise-floor level based on receiver bandwidth. It appears the author assumes that the reference is just to the thermal noise-floor level.
For example, in one reference the author cited the TIA102.CAAB performance recommendation, which refers to a static reference sensitivity of -116dBm at 5% BER. This value is correct. However, the author failed to include the noise figure of the receiver in his analysis. For example, a noise figure of 10dB would result in a required C/N figure of 10.4dB [-116dBm static sensitivity -(-136.4dBm thermal noise floor at 5.76kHz IF bandwidth +10dB noise figure)]. This figure is much closer to the 7.6dB value as shown in Table 3 on page 24. Furthermore, when one considers that TIA’s specification is a “minimum” value and that manufacturers’ equipment typically exceeds that figure by 2dB to 3dB or more, the Table 3 C/N value is confirmed.
In another example, the information from the government Web site referred to by the author is accurate, but the author’s interpretation is incorrect. The -121.4dBm level measured is for static sensitivity at 5% BER that occurs at 7.6dB C/N. This calculates to a receiver thermal noise floor of -129dBm, implying a receiver noise figure of 7.4dB (-129- (-136.4)). The published static sensitivity for that receiver is -119dBm, so that particular receiver exceeded the manufacturer’s published specifications by about 2.5dB.
The author also misconstrued manufacturer’s data. The level that was referenced (14dB) refers to the faded 5% BER, which is specified in TIA102.CAAB as being 8dB more severe than static conditions. Properly represented, this material would produce a calculated faded C/N of 15.6dB (8+7.6) relative to the receiver’s thermal-noise threshold. However, these should accurately be portrayed as “minimum” values, even though typical measured values are between 6dB and 8dB. That brings the manufacturer’s value of 14dB into the expected range.
We cannot support the author’s value of 6dB for analog FM. It lies somewhere between a static 12dB and 20dB SINAD. Under static conditions, it could be argued that this is a reasonable value. However, it would not be a representative value under faded conditions that delivers acceptable audio quality to the user. As indicated in Table 3, the recommended faded C/N value is 17dB, an 11dB difference before adjusting for filter bandwidth differences.
We are particularly disturbed that only anecdotal evidence was presented claiming lower range from Project 25 C4FM, while range was represented as an absolute derived from empirical data. Many narrow bandwidth digital systems have been installed, and coverage is always greater than analog unless there is interference present*.
Published industry reference
We would encourage you to review the correct criteria values when making your own assessments. Those values are located and published in “TIA/EIA Telecommunications Systems Bulletin TSB88-A, June 1999.” TSB88-A is a technical reference document developed in conjunction with the entire suite of Project 25 documents by a committee representing vendors, users, frequency coordinators and regulators. It was developed to provide frequency coordinators with the necessary information to determine frequency reuse as well as with a methodology for determining coverage range in the presence of co-channel and adjacent-channel interference. In that context, this information allows for the use of computer programs to evaluate potential frequency selections and to maximize coverage range while minimizing interference.
Because different digital modulations have different performance characteristics, a range of channel performances was developed, loosely based on the “circuit merit” definitions used for analog system evaluation. The definition is described as delivered audio quality and refers to:
- How easy or difficult it is to understand the intelligibility of a signal.
- Whether or not part of the audio is missing, requiring the message to be repeated, and if so, how frequently.
- What the detractors are, e.g., noise for analog or artifacts (distortions) for digital.
New systems are designed for a minimum DAQ of 3.0. A DAQ 2.0 is considered to be difficult to listen to, requiring considerable effort and frequent repetition for understanding. The higher DAQ values are defined for higher performance levels. They are defined as equivalent intelligibility to static SINAD values, but in the presence of Rayleigh fading (multipath), as experienced in actual usage.
Table 3, extracted from TSB88-A, shows the values for reference sensitivity (static C/N ratios) and the faded carrier to interference plus noise ratios, (Cf/(I + N)) for various DAQ values. The interference component, I, is included because interference is considered to be another detracting noise source.
It is important that a user or system engineer recognize that static conditions are only valid for bench measurements and are a way of comparing performance of individual units. By subtracting the Cs/N, the “noise floor” of the receiver can be determined. The Cf/(I + N) then determines the signal level required in the presence of fading and other potential detractors.
For example, analog 12dB SINAD Cs/N is 4dB. If a manufacturer’s published static receiver sensitivity is -119dBm, then the receiver’s “thermal noise threshold” would be at -123dBm. If faded analog DAQ 3.0 Cf/N is 17dB, then the faded signal level for system design would be -106dBm (-123dBm thermal noise threshold +17dB Cf/N).
The width of the IF filter also needs to be considered. Wider IF bandwidths produce less distortion and better channel performance at lower Cf/N. However, this comes with a price because wider IF bandwidths are more susceptible to adjacent-channel interference and result in higher thermal noise powers in the receiver.
The author selected 14dB for C4FM (equivalent to a DAQ 2.0 [in fading]), the static value (7dB) for π/4 DQPSK and a static value (6dB) somewhere between 12dB SINAD and 20dB quieting for analog FM. It is difficult to conceive how three disparate criteria could be used to arrive at a uniform and accurate conclusion.
Comparing digital modulations
As can be seen from Table 3, compatible four-level FM (C4FM) shows slightly degraded performance when compared to π/4 DQPSK, yet both have better performance than analog FM, particularly at higher Cf/N values.
As stated earlier, one should use a constant receiver noise figure reference. For the Table 1 analysis, the values were calculated assuming the same noise figure for all receivers. The comparison then becomes a simple matter of determining the difference due to bandwidth and the difference required for DAQ 3 and then calculating the relative ranges using a power exponent of 4.
The differences between noise power of various IF bandwidths are indicated in Table 4 on page 28.
The difference in faded C/N between analog FM (17dB) — digital C4FM (16.5dB) = 0.5dB. (0.5dB + 3.44dB for the difference in the required receiver IF filter bandwidth = 3.94dB.)
The range difference using analog FM as the reference is:
The difference using π/4 DQPSK is
Taking the absolute best case where the π/4 DQPSK uses a 5.7kHz ENBW filter would produce a 1.35:1 ratio, an 8% increase over C4FM, but at the price of a non-compatible over-the-air interface.
The author discussed the distortion effects of transmitter filtering on the modulation. Typical π/4 DQPSK implementations use a raised cosine pulse shape with the pulse-shaping split between the transmitter and the receiver. This technique, called matched filter, can be shown to result in theoretically optimum sensitivity under Gaussian noise. Part of the Project 25 deliberations was matched filters vs. pulse shaping. Project 25 decided not to select a matched filter system but rather to concentrate all the pulse shaping in the transmitter. Doing this makes the modulation compatible with linear and constant envelope (FM) transmitters and receivers. A slight loss of sensitivity exists compared to π/4 DQPSK. However, the selected modulations, C4FM and linear CQPSK, are compatible in a Project 25 receiver. Thus. the receiver does not care if the signal is modulated with C4FM or linear CQPSK. This allows a 12.5kHz channel bandwidth transmitter using C4FM modulation to communicate directly (over the air) with a 6.25kHz channel bandwidth receiver as long as the center frequencies are aligned. This creates a migration path to narrowband 6.25kHz channel systems. This migration path was encouraged and endorsed by the Project 25 user community.
Some of the author’s comments about how digital technology degrades over-the-air-performance need to be considered and put in the correct context. Digital Project 25 radios have significantly greater range than analog radios for systems designed for DAQ 3.0 or above. At the fringe of coverage (below signal levels normally used for system design), digital begins to break down at a faded BER of about 8%. This is in the general range of 12dB to 14dB Cf/N. At these weak signal levels analog would still be understandable (with great effort) while the digital radio would mute.
However, migration to narrowband analog FM does not improve this situation. Analog FM doesn’t perform as well in narrowband channels as it does in 25kHz channels. If narrowband analog is deployed, there is a 6dB degradation in performance from reduced deviation coupled with a 3dB improvement in receiver noise performance due to the narrower IF filter, resulting in a 3dB overall degradation. High-signal performance is reduced and a high SINAD cannot be achieved because some FM sideband information is lost passing through the narrow IF filter. Also, narrowband analog becomes more susceptible to noise pops, giving up the advantage that normal analog FM enjoys.
*Interference has been increasingly evident in portions of the 800MHz band where nearby cellular/SMR deployments create coverage holes due to their high signal levels where the desired signal is relatively lower. Both digital and analog systems are affected, but digital systems will mute while analog systems may provide audible clues to a user. The absence of audible clues is frequently reported as a range problem. This topic was the cover story in MRT‘s March 2001 Issue, “Why can’t we talk? It’s the interference.” Visit www.apco911.org for a copy of the Best Practices Guide described therein.
Muting thresholds: Stephen Bartlett’s response
I appreciate Mr. Olson taking the time to offer his perspective on my article “Does the Digital Radio Standard Come up Short?” (April 2001, MRT), and his candor in pointing out several key issues he had with it.
I wish to clarify to Mr. Olson, and the mobile radio community, a misunderstanding about what my article was reporting. The study I described wasn’t about the added range of audio quality that the P25 digital design achieves, but rather a discussion of the differences between the narrowband digital and wideband analog FM systems at the edge of radio coverage, or in other low signal-to-noise environments. As Mr. Olson pointed out, systems are not typically engineered for these fringe areas. However, it is under these conditions that radios will either mute or not, making coverage differences especially noticeable. The focus of my article was on how these different muting thresholds affect absolute range.
Mr. Olson is justified in suggesting the use of the number 4 for the power-law exponent describing the path-length dependency, in deference to the “extremely liberal” exponent of 2 that I chose. His figure reflects good engineering practice when designing mobile systems. It also reduces the scale of any range difference ratio.
He was correct to point out my not mentioning the receiver noise figure (NF) in the overall noise power (Pn) calculation. The correct formula is actually (neglecting interference):
For the analysis, I defined all receivers with equal and ideal NFs, but neglected to clarify this point. Where I erred was in assuming the manufacturer’s specifications, and other measured data I used to compare with the 4-FSK model, did the same. Many published specifications don’t clearly state whether their data are static Cs/N, faded Cf/N or if the receiver noise figures have been removed, included or normalized in their published values.
Muting thresholds will affect the coverage of the mobile radio system. Comparing these thresholds is the challenge. In his article, Mr. Olson stated that using a faded channel (Rayleigh fading) Cf/N ratio is more appropriate for modeling system performances than static Cs/N ratios. He also mentioned that no analog fading sensitivity standard exists, and he further recommended a faded value of 17dB for FM systems. This may be appropriate for determining a DAQ of 3.0, but not for a threshold mute. There are references showing critical wideband FM receiver thresholds are actually closer to 10dB; less with threshold extension (Rappaport, Wireless Communications). For digital systems, the mute threshold is defined by the critical bit-error rate where data cease to be reliably detected and vocoders to operate. It was good of Mr. Olson to recommend using the more realistic 2.6% BER threshold value for the IMBE vocoder to deliver more accepted audio in a faded channel. However, because of the focus of my article, I chose the 5% BER value as the digital mute threshold.
For a more complete approach in determining the expected faded channel performance for the digital systems, refer to J. Proakis’ book Digital Communications, Fourth Edition, where the ideal theoretical performances of many common digital modulation schemes in a Rayleigh faded channel are derived. In such a channel, the error rates vary inversely with signal to noise — unlike static exponential error behavior. In the case of binary, noncoherent, orthogonal FSK modulation, a Cf/N value can be found directly from the probability of error (Pb) using the Rayleigh channel formula: Cf/N (dB); 10log (1/Pb). We can use this equation to calculate the Cf/N performance for any FSK signal at 5% BER. Orthogonal binary FSK and 4-FSK channels will perform nearly identical except at low Cf/N. Therefore we can use the Rayleigh formula to estimate C4FM’s average departure from ideal performance in weak signals due primarily to its nonorthogonal four-level modulation.
Mr. Olson reported that in the faded channel, C4FM will break down in the fringe areas with 8% BER at a Cf/N of 12dB to 14dB. The Rayleigh channel formula predicts the ideal binary FSK Cf/N value at 8% BER to be 11dB. Thus, Mr. Olson’s 8% BER value shows that C4FM has from 1dB to 3dB of degradation (2dB average) from ideal. From this, we can determine the optimal Cf/N mute threshold for C4FM modulation at 5% BER:
Using this optimal C4FM mute threshold of 15dB, the referenced wideband FM mute threshold of 10dB, and the bandwidth adjustment of 3.44dB, an absolute range difference between the digital C4FM and wideband FM analog systems can be calculated, using the equation recommended by Mr. Olson:
This 0.91 value shows a potential loss of absolute range between the C4FM digital and wideband FM analog systems, due to their different muting thresholds. Realistically, in the unpredictable faded channel with the parameters of analog and digital modulation being so different, there may be a variety of cases where either system may outperform the other, perform nearly the same or vacillate between these extremes — especially in fringe areas where muting occurs. It is in the best interest of the community to know how these digital narrowband systems perform in a variety of locations and scenarios as they become more widely implemented. If problems can be predicted early, they can be mitigated early.
— S. Bartlett