Making time, holding time with or without a GPS.
The most ubiquitous way to get accurate time and phase, every time, everywhere is to use a GPS disciplined oscillator (GPSDO). Over the last few years, some concerns have arisen due to the potential impact on critical infrastructure of malevolent attacks such as jamming or due to natural causes such as solar flares.
The GPS system has been designed to distribute time around the world. It gets time from a reference called a time scale. They are several time scales in the world. In the US, NIST and USNO are used for the UTC time scale and are referred to as UTC (NIST) and UTC (USNO). USNO also produces the GPS time scale. The time scales around the world are coordinated through an international agency called BIPM. USNO, NIST are all coordinated to BIPM through a mechanism called two-way time transfer.
BIPM defines International Atomic Time (TAI) from ensembling atomic clocks that are disseminated around the globe - the duration of the TAI second is kept as close as possible to the SI second. Coordinated Universal Time (UTC) is an atomic time scale derived from TAI to provide a reference scale in step with the irregular rotation of the earth. As a result of this process, the mean solar day, which is nominally 86400 seconds long, is actually getting longer when measured in SI seconds with stable atomic clocks.
It is the role of IERS (International Earth Rotation Service) to observe the self-rotation of the earth and to make sure that when the earth completes its rotation - 24 hours have elapsed…. It does the same with respect to the earth’s rotation around the sun. As the earth slows down due to tidal effects, leap seconds have been used to adjust the time periodically since the original 1/1/1970 00:00:00 epoch.
TAI is 35 seconds ahead of UTC today (dynamic value). TAI is always ahead of the GPS timescale by 19 seconds (fixed value).
In conclusion, atomic clocks are an integral element of any time scale. GPS/GNSS systems are not the only means to distribute time. Many forms of two-way time transfer techniques are being deployed today over a dedicated network, over satellites (the Ku band is generally being used) and, to some extent, over non-dedicated networks. In the latter case, some careful engineering is required as time transfer over a non-dedicated network is very sensitive to network asymmetries. (see ITSF 2013 presentation on sources of asymmetries in a fiber network). The PTP protocol as defined by ITU uses the TAI timescale whereas NTP uses UTC. These are two mechanisms of time transfer over a non-dedicated network assuming that there are no asymmetries (or that asymmetries are externally compensated).
Why frequency, phase and time matter in cellular networks.
Frequency Division Duplex (FDD) has been the dominating Radio Access Network (RAN) technology since the inception of cellular networks. It requires paired frequency bands - one for transmission and another one for reception. Time Division Duplex (TDD) uses a single frequency sharing the channel between transmission and reception and spacing them apart by multiplexing the two signals in real time. Stable, known, and accurate frequencies are fundamental to enable user equipment (UE) to connect to base stations. Worse, Doppler shift due to UE in a train or in a car will render impossible any communication unless the UE can syntonize the receiving frequencies from nearby base stations. In both cases, FDD and TDD, without this capability, UEs will be competing for the same frequencies or time slots without being aware of wrongdoing and potentially rendering the cellular network inoperative.
The accuracy of the radio frequencies being used is standardized and measured in parts per billion. A frequency stability of ± 50 ppb corresponds to a drift of ±4.32 ms per day. These frequencies are created by local oscillators within the UE and the base stations. The UE’s inexpensive local oscillator is syntonized to the radio frequency it is receiving from the cellular network. By being able to syntonize its local oscillator to the received signal coming from distant base stations, the UE can demodulate the receiving signals, decode the protocol, extract information from the payload carried by the protocol over the control plane, answer requests from the network and eventually send and receive data over the data plane.
The base station oscillator is more expensive as it has to have a very low drift (± 50 ppb). However, even this drift needs to be corrected. Depending of the quality of the oscillator and the requirements in terms of stability and accuracy, a reference or master signal is required. Traceable T1 or E1 feeds can be used as well as network protocols like NTP or SyncE. In the case of NTP, the local oscillator must be of better quality and be able to maintain its stability over a large gradient of temperatures. When using either an E1 or a T1 traceable feed as well as SyncE, the base station oscillator is syntonized to the high quality frequency extracted from the receiving signal. PTP (aka the 1588 protocol) can be also used to re-synchronize the local oscillator – the PTP protocol can support less expensive oscillators than NTP and still keep that local oscillator within the requirements due to the higher packet rate. By traceable, we mean a feed that is locked to a high quality clock (Cesium clock, GPSDO,..). Remote NTP server clocks and PTP master clocks use GPS-disciplined oscillators (GPSDO) to prevent their local oscillator from drifting.
TDD requires phase alignment between UEs and cells within the radio access network in order to avoid inverse link interference. As TDD is starting to be deployed over FDD due to its better overall spectral efficiency and the flexibility to adapt UL and DL bandwidth allocation, new deployments are moving from frequency only RANs (FDD) to frequency and phase aware RANs (TDD). When frequency and phase or time are required at the base station, then either a local GPS receiver is used and/or PTP. The NTP protocol typically can’t convey phase information with enough accuracy (NTP can carry time and phase within a sub millisecond accuracy, however, most radio networks require sub 10s of microseconds and for LTE, phase requirements are within the microsecond range).
When phase requirements are not met within the RAN, several problems can happen occur even if frequency requirements are being met. Handover from one base station to another one may not work when an access method like TDD is being used. Worse, a UE can usurp the time slot allocated to another UE (UL signals are using SF single-carrier frequency division multiple access (SD-FDMA mode)) and then can disrupt or scramble the communication.
In dense areas where LTE is being deployed, inter-symbol interferences will prevent UE’s from demodulating the DL orthogonal frequency division multiple access (OFDMA) signals from neighboring LTE small Cells (LTE base stations), rendering the network useless and wasting the capacity of the network or worst case exhausting the spectrum as the UEs will then try to connect to the remote congested macro cell instead of the neighboring small cells.
LTE Advanced requires Frequency Coordination between the small cells and the macro cells in order to mitigate interference and to boost coverage and capacity. In order to reduce delay, to minimize TDD UL/DL switching latency, to improve UL and DL available bandwidth from the small cells to the UEs by using carrier aggregation and MIMO antennas, macro cell/small cell coordination techniques have been introduced and are being leveraged within the LTE RAN. In these cases, not only phase alignment is required between the small cells and the macro cells but also absolute time accuracy. (see 3GPP Rel 12).
Where historically cellular has been dominated by FDD, over time we can expect TDD to dominate the radio network landscape as TDD is not only a technique that provides better spectral efficiency, dynamic bandwidth allocation (DL/UL ratio) but it is also the technique that maximizes the use of un-paired frequency bands and also is supported for roaming between carriers. Frequency requirements for FDD and TDD are similar and easily met with modern assistance techniques like PTP, NTP or SyncE. Frequency and phase requirements need to be carefully studied and planned when deploying HetNet RAN. GPS carries a huge installation cost and the antenna cable length has to be calibrated. Furthermore GPS is prone to jamming and spoofing and frequency assistance is required in order to defeat this threat. Most ITU and ATIS recommendations are in favor of PTP as the solution for phase and frequency.

Some Q&A
What is the difference between “spoofing” and “jamming?” 1) They are both illegal! 2) Jamming consists of sending a strong signal that saturates with noise the band being used by the satellites for broadcasting data to the GPS receivers on Earth. This is a very basic attack. Holding time can be done by the local oscillator or by some sort of PTP assistance that offsets the drift of the local oscillator if the jamming lasts. 3) Spoofing is the faking of a rogue / fake satellite that makes the GPS receiver believe that it is not where it should be. Several techniques can be used to counter such an attack. First, a base station doesn’t move so any detection of a displacement should be considered as an alert. Second, a stable oscillator such as an atomic clock will prevent a rogue attack on the time (see http://www.army.mil/article/88361/Miniaturized_atomic_clock_to_support_Soldiers_in_absence_of_GPS/).
How does a GPS receiver decode times from satellites? 1) A GPS receiver receives signals from 4 different satellites (at least) – there are 4 unknowns - the coordinates: which are the geodetic coordinates x,y,z and the time t. 2) GPS satellites orbiting around the earth complete a rotation in half a sidereal day (a little bit less than 12 hours), broadcast a frequency reference f, time, and information related to their positions/trajectories. 3) As satellites are rotating around the Earth, the frequency reference f is Doppler shifted. Once the GPS signals are isolated from the noise, the GPS receiver locks to the frequency reference of each satellite, decodes the data and then starts to process all data that are broadcasted by the satellites in view. 4) In order to get an accurate position and an accurate time, the GPS receiver has to cancel out the propagation time from the satellites through the ionosphere and atmosphere and through trilateration, in order to identify where on earth it is located. As Earth is not perfectly spherical, a model of Earth is being used to do a correct assessment of the location.
How does a time scale work? 1) To make time, you need a very stable frequency. Atomic clocks such as Maser sources, Cesium sources or Rubidium sources provide very stable frequencies that are locked to atomic resonances. 2) For greater stability, these sources are ensembled – the resulting source will have a factor of improvement that is proportional to the square root of the number of sources being ensembled. 3) Then the system is initialized to a reference (can be BIPM or even a local accurate GPS receiver once in a while) and then is able to keep time in a free-wheeling mode for a very long period of time (months, years) with a very high accuracy.
How LTE FDD and TDD differ? 1) FDD used paired frequencies, one for transmission and one for receiving. TDD is using a single frequency for transmitting and receiving 2) Uplink (UL) and downlink (DL) are divided into radio frame, each 10 ms in length. The frame consists of two "half-frames" of equal length, with each half-frame consisting of 10 slots or 8 slots plus 3 special fields: the downlink pilot time slot (DwPTS), the guard period (GP) and the uplink pilot time slot (UpPTS). Each slot is 0.5 ms in length and two consecutive slots form exactly one subframe. The length of the individual special fields depend on the uplink/downlink configuration selected by the network, but the total length of the three fields remains constant at 1 ms. This resource structure is the same for both TDD and FDD. 3) TDD doesn’t use guard band between paired frequencies but a guard interval of time to allow the switching between DL and UL