Final Report from USCGC Healy

Warm-Water and Cold-Water (Leg 1) Science Testing of

150kHz and 300kHz Broad Band Vessel-Mounted ADCP Systems


2000/7/18

Jules Hummon, Eric Firing

University of Hawaii


Executive Summary


The USCGC Healy was delivered with two shipboard ADCPs installed: RDI BB-150 and BB-300, where the number is the nominal operating frequency in kHz. After two test cruises, the BB-150 appears to be operating normally, but the BB-300 is not. We have not determined the cause or causes of the BB-300 failure. However, based on the characteristics of ADCPs as a function of frequency and the performance of the BB-150 on the Healy, we believe there is no useful scientific role for the BB-300 anyway, and it should be removed.


The primary factor limiting scientific usefulness of a modern shipboard ADCP system is its effective profiling range. Compared to ADCPs on most other US research ships, the range of the BB-150 on the Healy is poor: at most 250 m in these trials, under ideal conditions. There are three reasons: (1) Absorption of sound by the acoustic window. To some degree this problem is unavoidable on an ice breaker. We do not know whether the problem is any worse on the Healy than on other ice breakers. (2) Sound absorption by a layer of bubbles swept under the transducer. This problem depends on speed and sea conditions. Under moderate conditions, it renders the Healy's ADCP useless at speeds over about 14 kts. (3) The "broadband" implementation in the BB-150 inherently sacrifices range for short-term accuracy, which is not really useful on a shipboard system.


Interestingly, under no conditions did we find a major increase in background acoustic noise with increasing speed; apparently neither flow noise nor propellor cavitation noise is a significant problem with the ADCP on the Healy.


Interference, whether acoustic or electric, was a primary focus of testing. There was no evidence that the BB-150 interfered with other sonars. The 3.5-kHz and 12-kHz depth sounders caused only minor interference with the BB-150. The only serious interference with the BB-150 was from an as yet unidentified source. When present, it affected every third depth bin, and caused serious velocity bias. It reached the ADCP from inside the ship, not through an external water path, but we don't know whether it was acoustic or electromagnetic.


With the 4-m and 5-m blanking intervals used in testing, weak reverberation (ringing) biased velocities in the top depth bin only. We did not test longer blanking intervals, but we expect that no more than 13 m would be needed to eliminate the ringing; smaller values should also be tested.


Wave-induced oscillatory vertical motion of the transducer causes velocity ambiguity wrap-around to occur at much slower forward velocities than in the absence of vertical motion, so it is essential to use a large ambiguity interval when the ship is underway. Therefore the standard operating mode must be WM1, WB0, WV650; WB1 will cause loss and/or corruption of measurements.



Introduction


Initial testing of the two shipboard acoustic Doppler current profilers (ADCPs) on the USCGC Healy was carried out during two cruises designed to evaluate science systems on board. The first cruise ("warm water trial") took place February 19-28, 2000, between Pensacola, Florida and San Juan, Puerto Rico. The ADCP testing was predominantly in deep water. Seas were mostly calm. As is typical in the blue waters of the tropics, the acoustic backscattering was weak. The second cruise was leg 1 of the "cold water trial", May 25-31, 2000, with both ports of call being St. Johns, Newfoundland. Testing included shallow water over the continental shelf as well as deep water north of Flemish Cap. Conditions during this cruise included some moderate swell (8') and brisk winds (10-40 kts, but rarely over 30kts). Acoustic scattering was markedly stronger, as expected for subpolar waters.


Testing was designed to evaluate the performance of both shipboard ADCP systems. Issues addressed included acoustic interference from other sonar devices, effective profiling range as a function of speed and sea state, sonar accuracy, and the quality of the navigational data inputs. The latter was primarily a problem during the warm water trial; several problems were fixed prior to the cold water trial. Most of the diagnostic analysis of sonar performance was done by looking at single pings, often in beam coordinates (as opposed to geographical coordinates). Hence we were able to see phenomena that would have been masked by the averaging that is routinely done prior to standard processing.


This report begins with a description of the shipboard ADCP systems; additional background information and explanation of jargon is provided in the Appendix. Details of the tests and results follow, and we close with recommendations. Figures are in here


System Description


A shipboard ADCP (acoustic Doppler current profiler) system is comprised of several components: a sonar, navigation sensors, and a data acquisition system. The sonar includes a four-element transducer mounted in a well on the bottom of the ship, a deck unit that generates the acoustic pulses and processes the returns, and a 41-wire cable to connect them. Essential navigation sensors include a GPS receiver to provide position fixes, the ship's gyrocompass to provide reliable but inaccurate heading, and an Ashtech 3DF GPS attitude sensor to provide accurate, but not perfectly reliable, heading. The data acquisition system consists of a PC and a data acquisition program (DAS) that controls the sonar, integrates data from the sonar and the navigation sensors, and records the data, usually after transforming and vector-averaging the velocities.


On the Healy, there are two independent ADCP data acquisition and sonar systems, one operating at 153 kHz (BB150) and the other at 307 kHz (BB300). Both sonars are Broadband (coded pulse) units made by RD Instruments (RDI). In contrast, most United States research ships use narrow bandwidth (uncoded pulse) sonars, also made by RDI.


The two transducers are mounted in separate wells or seachests in a small sonar room called the "transducer well". Other seachests house additional sonars, including the 12/3.5kHz Knudsen (The Bathy 2000 can also run these transducers), the 33 kHz Bathy 1500 and the 350 kHz Dual Axis Doppler Speed Log (SRD421S). The transducer well is located near frame 60, about 1/4 of the ship's length aft of the bow. The nominal "forward" axis of the BB150 transducer is oriented 54.4o to starboard. (A more typical orientation would be 45o; we suspect the actual orientation is accidental.) The sonars are separated from the sea by a thick acoustic window, which protects them from ice damage and permits them to be surrounded by a fluid other than seawater. The acoustic windows in front of the BB150 and the BB300 are "a proprietary material called 'SeaBeam Orange', a type of polyurethane" according to Dave Kilonsky, an Avondale engineer.


For each BBADCP data acquisition system, a 200-MHz Pentium computer with a 2-GB disk running Windows95 was set up with a dual-port RS-422 serial interface board in addition to the usual pair of built-in RS-232 serial ports, allowing the PC to handle up to four serial inputs.


The choice of DAS software was between two RDI programs: Transect, which has been around for many years, and VmDas, which is still in development. Transect runs under DOS, VmDas under Windows95/98/NT. Data acquisition was primarily done using Transect for the warm water trials, but VmDas and Transect were used alternately with the BB150 during the cold water trials. Transect can log serial navigation inputs only from a single port, whereas VmDas can log two inputs. To facilitate switching between Transect and VmDas, only a single port was used.


Two types of navigational data input are critical to a shipboard ADCP system: position and heading. The most accurate position fixes are from a P/Y-code (military) GPS receiver or, when available, a real-time differential receiver. Heading is comes from two sources: reliable but inaccurate heading comes from the ship's gyro compass, and less reliable but much more accurate heading comes from an Ashtech 3DF GPS attitude sensor. Because of its unreliability, the latter must be used in postprocessing, not as a sole source of real-time heading. The preferred way of obtaining the gyro compass heading is via the synchro interface on each ADCP deck box. These inputs were not available until the the synchro cable was rewired during the cold water trials. Additional details of the navigational inputs will be given in the following section.



Instrument Evaluation


150kHz BBADCP



Performance


The main performance parameters for a shipboard ADCP system are depth range and accuracy. Both depend strongly on external factors, particularly noise (electrical and/or acoustic), the presence of bubbles or ice beneath the transducer, and the presence of acoustic scatterers in the water. Of these, only the latter is independent of the ship and installation; noise and bubbles vary with ship and transducer well design, sea state, and what the ship is doing. Steaming at full speed into a head sea, for example, usually maximizes both acoustic noise and bubbles.


For scientific applications, shipboard ADCP data are normally vector-averaged in time, typically in 300-second ensembles, then combined with navigation data to yield water velocity over the ground. The effective depth range of time-averaged and navigated velocities is less than the maximum range of individual pings; it is the maximum depth at which a substantial fraction of pings in each ensemble have valid velocities. The minimum acceptable percentage depends on several factors; it can be as low as 30% at the bottom of a profile provided the single-ping velocities are vector-averaged in two stages, with the shears averaged separately from the velocity in a reference layer. The old narrowband DAS 2.48, and the new VmDas can do this kind of averaging; Transect cannot, so a high percent good threshold must be used uniformly with Transect, reducing the effective range. It is important to have percent good as high as possible, typically at least 80%, in a range of shallow depth bins, so that the ensemble average of the velocity in this reference layer is a good approximation to a true average of the velocity relative to the ship.


The depth range of the BB150 during the cold-water trial was, in general, substantially greater than during the warm water trial. Under the best conditions of the warm water trial, the effective range of the BB150 was 160 m; under typical conditions, it was 110-145 m. Data were collected during the warm water trials in relatively calm seas. In the cold water trials, under the best conditions, the deepest range seen was 250 m, with more typical good ranges of 150-220 m (Figure 1). Conditions sampled included some which were sufficient to render the data useless (bridge-reported swell of 8', seas at 2', ship speed 16 kts, seas and swell on the port beam). Data range decreased dramatically at speeds over 13.5kts even in calm seas. Other factors detailed below decrease the effective data range.


For comparison with another widely-used ADCP, a narrowband RDI NB150, the system on the Antarctic icebreaker Nathaniel B. Palmer had an effective range of up to 230-260 m under good conditions on a transit of the tropical Pacific in 1998. In high latitudes, profiling frequently reached 400 m under good conditions. Although direct comparisons have been rare, all past experience as well as theory lead to the expectation that a BB150 will have substantially less range than a NB150. This BB penalty can be reduced somewhat by using setup parameters other than the ones we chose; unfortunately, the setups that maximize range (primarily WM7) are not suitable for routine use underway.


Four distinct types of noise have been found in the BB150 data: bubbles, ringing, acoustic interference from other sonars, and a fourth (as yet unidentified) source of interference. Ice is expected to strongly and adversely affect data quality, but ice conditions were not sampled during leg 1 of the cold water trials.


On many shipboard ADCP installations, the background noise level--the noise floor--rises sharply with sea state and with the degree to which the ship is steaming into head seas, entraining bubbles into the boundary layer flowing beneath the transducer. We do not know how much of this noise is caused by bubbles at the transducer faces, and how much is caused by increased propellor cavitation, which is often the main source of underway noise. On the Healy, however, the increase in the noise floor was almost imperceptible; the degradation in performance coincident with bubble layers under the transducer was almost entirely caused by absorption of the outgoing ping and the returning signal.


To test the ADCP performance as a function of ship's speed, all other sonars were turned off, and ADCP data were collected while the speed of the ship was increased in discrete steps from 1 kt to over 16 kts (Figure 2). There was little swell during the test, and some of the best depth penetration of the cruise were obtained. Although depth range did not decrease monotonically as the speed increased (part of the variability was caused by ocean conditions), there was an obvious major loss of range, and eventually of a large percentage of entire pings, at speeds above 12 kts. The loss was correlated with the autopilot-induced oscillations in heading (Figure 3), which were as large as 10o to either side of the intended course. At the highest speeds, the single-beam signal strength shows that single pings, and sometimes short groups of pings, were completely blocked by the bubble layer (Figure 4).


Acoustic reverberation ("ringing") of the ADCP ping (see appendix) can be seen as reduced correlation in the top bin in Figure 3, and more importantly, as a top bin bias toward zero velocity relative to the ship in Figure 6. Ringing probably contributes to the northward (and to a lesser extent the eastward) velocity component in the top bin in Figure 1. Overall, our impression is that the ringing problem on the Healy is relatively mild. Ideally we would have run a set of tests with increasing blanking interval to find the minimum value that eliminates ringing, but this has not yet been done. Because we have detected no sign of ringing in the second depth bin, we suspect the blanking interval would need to be no larger than 13 m, and probably less. Only values of 4 m and 5 m were used during the trials.


Acoustic interference from other sonars is not a major problem (Figure 5). Neither the Seabeam nor the Doppler speedlog caused any perceptible signal in the BB150 measurements. The Knudsen 12 kHz and 3.5 kHz depth sounders caused minor interference, typically in three adjacent depth bins at a time, lowering the correlation when combined with backscattered sound (within the depth range of valid data), and causing slight elevation of correlation at depths beyond the profiling range. In shallow water, interference seemed to come from reflected energy as well as from the depth sounder pings. The interference can cause isolated spikes in the velocities, but after screening for error velocity and vertical velocity, the effect on ensemble-averaged velocities is minimal. The BB300 interferes more strongly with the BB150, but because we don't expect the BB300 to be used, this is not a problem.


The most serious type of noise received by the ADCP has well-defined properties, but its source has not yet been identified. We refer to this noise as "hash" because of the way it looks on a correlation plot such as Figure 7. It shows up as high correlations and amplitudes in a single depth bin of all four beams simultaneously, at three-bin intervals corresponding to 0.037sec. Pings free of the interference occur at intervals of several seconds, but the interval was not constant from one instance to another. In a warm-water example, there where whole pings in groups of 1-3 with no hash, separated by several hashed pings; in a cold-water example, nearly every 6th ping was devoid of hash (i.e. every 11 seconds, the hash is missing). Other factors such as the presence of bubbles, operation of other sonars, or ship's speed, pitch, roll, or heading did not not consistently correlate with the presence or strength of the hash. In many cases, the hash stopped when the ship slowed down (Figure 7). However, an example with particularly high correlations occurred during the cold water trials at only 2.2 kts (Figure 8, region A).


Figure 8 illustrates the two main problems--bubbles and hash--in three typical data samples from the cold water trial. Conditions are summarized below:


seg --roll--- --pitch-- sog ship wind swell sea wind swell sea

mean std mean std kts head dir dir dir kts ft ft

A -3.1 0.9 -0.1 0.5 2 320 -50 -150 -60 10-15 4 1

(between A and B) +50 -150 +40 20-25 5 5

B -0.1 0.6 -0.1 0.5 11 330 0 -100 -30 0 5 2

C -0.8 0.8 -0.1 0.5 15.5 290 +50 0 0 20-25 6 2


Panel A (at 2.2 kts) shows the hash as high correlations, seen most clearly below 150m. Panels B (at 12 kts) and C (at 16kts) show much lower hash correlations, but B has some signal degradation due to bubbles, and segment C is devoid of useful velocity data, also due to bubbles.


The presence of hash in Figure 8C and Figure 9 shows that it is not coming to the ADCP transducer from the sea; if it were, it would be stopped by the bubble layer. Beyond that, there is little we can say about its source. It could be either electromagnetic, perhaps coming in through the transducer cable, or acoustic, coupled to the transducer via the hull.


The high-correlation hash (such as in Figure 8, panel A) can corrupt the velocity data because it causes systematic bias in all four beams (Figure 10). Screening the beam velocities with a higher correlation threshold is relatively ineffective, eliminating too much valid data for the amount of hash it removes (not shown). A strict error velocity screening criterion is better, but still inadequate. In addition, a vertical velocity criterion is needed (Figure 11). Given such screening, most of the bias can be eliminated, but only with the loss of a major fraction of the valid data as well.


Problems:


installation:


The transducer well labeled 150 appears to leak (1 gallon lost in 15 minutes). The pipes for the BB150 and BB300 were traced from the sight tubes to the transducer wells and no water appeared to be leaking from the pipes. The BB150 transducer well shows signs of previous leakage around the lid: rust dribbles down the sides are apparent. However, the amount of water on the deck (in the transducer well room) did not seem to match the amount of water used to refill the BB150 tank up to that time. We concluded that the leak is inside the transducer well, probably out the seal around the window. The expected source for refilling the tanks is fresh water. Continued refilling (topping off the tank) was not recommended because of the continued dilution of the remaining antifreeze in the tank.


velocity ambiguity:


Because profiling range decreases with increasing bandwidth, the medium bandwidth setting (WB1) is sometimes recommended over the widest bandwidth (WB0). Unfortunately, the maximum ambiguity velocity for WB1 is only 3.3 m/s. This velocity (along the beam) is often exceeded even at moderate ship speeds because of pitch and heave.


The argument sometimes given for the minimum acceptable ambiguity velocity, applied to the Healy (transducer oriented 54o from the keel, 30o beam angle from the vertical) is that a forward velocity of 14.6 kts causes a maximum beam velocity of 3.3 m/s. Although the Healy is a remarkably stable ship, its wave-induced vertical motion under moderate conditions is sufficient to cause ambiguity wraps at speeds far lower than the calculated 14.6 kts (Figure 12).


Beam 1 wraps most frequently, in part because of its axial alignment, and in part because of its vertical alignment. Most of these wraps affect both vertical and error velocity, but in some the error velocity is small. This is another case where a vertical velocity data screen is helpful. A better solution is to use WB0 and its maximum ambiguity velocity command, WV650.



serial strings to the DAS computers:


At the start of the warm water trials, navigation and heading inputs to the ADCP came not directly from the sensors, but from the Voyage Management System (VMS), an integrated navigation system. Positions were provided not via $GPGGA messages from the receiver, but in similar $NVGGA messages, with poor precision and missing fields. The heading available to the shipboard ADCP was "switched heading", which could be either gyro or Ashtech heading: the type of heading provided at any given time was not specified. No genuine Ashtech message was included. In addition, the cable from the VMS to the BB150 was noisy, resulting in garbled messages. This system is still available, including the bad cable, but should never be used.


During the cold water trials, a direct serial line from the Ashtech 3DF GPS attitude sensor was run to the BB150 PC. While the Ashtech input is essential for correcting heading in post processing, it is not an ideal source of position fixes; GPGGA messages from a P-code receiver should be provided. Until this is done, science users of the ADCP should ensure that the P-code fixes are being logged by the ship's Science Data Network.


300kHz BBADCP:


During the warm water trials, the BB300 did not acquire water velocities below the first bin. The likelihood seemed high that the transducers were at least partly exposed to air. Since then, Marine Science Technicians on board the Healy concluded that this was not the case because they saw water escape when they cracked open a valve on the transducer well cover. They concluded that the transducer heads (which are below that level) are covered. There was no change in BB300 performance (or lack thereof) between the warm water trials and the cold water trials. In the remainder of this section we will describe what was done with the BB300 system, what was learned, and what remains unknown.


During the cold water trials, there was no functioning direct (synchro) input from the ship's gyro compass to the ADCPs. Prior to the warm water trial, the synchro cables were rewired appropriately. Although the new cabling and the BB300 gyro interface board appeared to be working, gyro heading was still missing from the raw data output by the instrument. Darryl Symonds suggested that the next diagnostic step should be to check whether the BBVM300 Electronics Chassis Backplane board has a fault in the RS485 bus; but the cold water trial ended before this could be done.


At present we do not know the acoustic loss caused by the window separating the transducers from the sea, so we cannot determine the contribution of the window to the poor performance of the instrument. Acoustic loss is expected to be a problem with both instruments, but the BB300 should suffer a higher loss than the BB150 due to its higher frequency.


Navigation and heading inputs to the BB300 are still coming from the VMS-supplied serial line, with its inherent problems and limitations. This problem could be solved as it was for the BB150.


Tests during the warm water trial showed that the 350kHz Doppler speed log completely obliterates the 300kHz BBADCP velocity signal. We suspect this would be true even if the BB300 were profiling to a reasonable depth.


In summary, the BB300 is not functional for two reasons: (1) there seems to be a fault in its handling of synchro heading input, and (2) it returns nothing at all below the first depth bin. We do not know the cause of either of these problems. Even if they were solved, interference from the 350 kHz speed log would present a major impediment to use of the BB300. Furthermore, the BB300 could at best achieve about half the poor depth range of the BB150. We can imagine no scientific application for which such poor range would be acceptable, or in which the BB300 would be preferred to the BB150 on the Healy.


Recommendations



  1. For most scientific applications of shipboard ADCP systems, the limiting factor is indeed range. A NB150 is marginal, and a BB150 behind a window is simply inadequate. The BB300 should be replaced with a system that gets at least the range of a NB150, and preferably much more. This may require waiting a year or two for newer low-frequency instruments to mature.


  1. Identify and remove the vertically-independent high-correlation signal which sporadically corrupts the BB150 data.


  1. Ambiguity wraps occur at speeds as low as 11kts using the setting of WM1, WB1, which RDI sometimes recommends. Therefore a setting of WB0 and a higher ambiguity setting (such as WV650) should be employed for routine use.


  1. The prevalence of the unidentified hash in the signal means that a vertical velocity threshold must be used in addition to the error velocity threshold to screen data before it is averaged. The vertical velocity threshold is only available in RDI's newer acquisition software, VmDas. We hope VmDas is stable enough to be reliable: it has many advantages over Transect. VmDas can use reference layer averaging, has vertical velocity editing criterion, allows 2 serial inputs, and is more intuitive.


  1. In addition to the genuine Ashtech heading string being sent to the DAS PC, pure P-code GGA messages need to go from a P-code receiver to the DAS PC. If necessary to provide a P-code GPGGA message stream for science use, a dedicated P-code receiver should be obtained--they are not expensive. VmDas can handle navigational inputs from two ports, but Transect has only one input; if Transect must be used, then a buffered serial port combiner should be installed to merge the P-code and Ashtech data streams on a line-by-line basis.


  1. A sound speed sensor should be installed in the transducer well, and its output logged so as to provide easy access for postprocessing of the ADCP data. This is essential because the calculation of velocity from Doppler shift depends on the sound speed at the transducer. Sound speed depends on the temperature and the composition of the fluid; although the temperature is measured at the transducer, the composition is in general not known with adequate accuracy (if at all), and we have been unable to find formulae for soundspeed in antifreeze mixtures.


  1. If not already in place, acoustic damping material should be used to line the transducer well, to reduce ringing. Until then, blanking intervals from 6 to 13 m should be tested to find the minimum necessary to eliminate ringing in the top bin.


  1. Whenever the transducers are removed (for maintenance) and reinstalled, maintain the present orientation.



Appendix: Instrument Background


To understand our analysis of the shipboard ADCP performance, and our recommendations for future use and improvement, it is helpful to review some ADCP principles and terms. The following is a minimal description of the RDI Broadband sonars on the Healy.


Shipboard ADCPs use four narrow beams of sound, each directed at an angle of 30 degrees from the vertical (this is the most common angle, and the one used on the Healy's instruments; other angles are possible), and arranged at 90-degree intervals of azimuth. A brief sound pulse is transmitted simultaneously by all beams, typically about once per second. As each pulse travels along the beam axis, a minuscule amount of sound is scattered by objects in the water, mostly zooplankton, and some of this backscattered sound returns to the transducer, where it is received. The frequency of the received sound is Doppler shifted in proportion to the axial component (that is, the component along the beam) of the velocity of the scatterer relative to the transducer. Measuring the Doppler shift as a function of time passage since the pulse therefore provides an estimate of one velocity component for each beam, as a function of depth. One second after a ping, for example, sound returns from 650 m depth, which is 750 m along a beam axis.


Now, divide the four beams into two pairs: for example, one pair might be directed fore and aft, the other starboard and port. Given the crucial assumption that the velocity of the water, and hence the acoustic scatterers, is roughly uniform in any horizontal plane intersected by the beams, the horizontal separation of the beams can be ignored, and their velocity estimates at a given depth can be treated as if they were made at a single location. The two axial velocity components can then be transformed into one horizontal velocity component (e.g., along the keel) and the vertical component. The second pair of beams gives an estimate of the orthogonal horizontal velocity component (e.g., athwartships), and a second estimate of the vertical component. Let the average of the two vertical estimates be the official estimate (w); the difference is then a measure of the consistency of the measurements among the four beams. RDI defines this "error velocity" (e) with a scale factor to make it a direct estimate of the instrument noise expected in a horizontal velocity component: e = (w12 - w34) * cot(q) / sqrt(2), where w12 and w34 are the vertical velocity estimates from beam pairs 1-2 and 3-4, and theta is the vertical beam angle.


Given the two horizontal velocity components (in "instrument coordinates") and the orientation of the transducer, the velocity vector can be rotated into east (u) and north (v) components. (See the document "ADCP Coordinate Transformation" available from RDI.) Vertical velocity (w) and error velocity (e) remain the same. This geographical coordinate system is often referred to as ENU, for east-north-up. In this description we have neglected tilt--we have assumed the instrument's vertical axis is indeed vertical. This is never exactly true, but it turns out that neglect of tilt in shipboard ADCPs has only a small effect on the estimates of horizontal velocity. There is a larger effect on the estimate of vertical velocity; for example, if the transducer is tilted up toward the bow by one degree, there will be an upward bias in w of 1.7 cm/s per m/s of ship's forward speed relative to the water. This is not a problem, because we know that the actual vertical velocity in the ocean, averaged over several minutes, is too small to measure--so at best, w, like e, can be used only for data quality evaluation. (The actual reason that tilt is rarely used in the transformation from instrument coordinates to ENU is that reliable tilt measurements are rarely available.)


The transformation from instrument coordinates to ENU requires an accurate measurement of the transducer heading. This is the sum of the ship's heading and the orientation of the transducer relative to the ship. The latter is fixed for a given instrument installation and is most often chosen to be plus or minus 45 degrees, so that two beams point 45 degrees port and starboard of the bow, and the other two point 45 degrees port and starboard of the stern. When the ship is underway, all four beams then measure Doppler shifts of similar magnitude: toward higher frequency in the forward-facing beams, and toward lower frequency in the aft-facing beams.


Estimates of Doppler shift, and hence velocity, are made at regular time intervals following each ping. The time intervals correspond to depth intervals, which RDI calls "bins". Immediately following the ping, there is a delay before sampling of the return signal begins. This delay, corresponding to a "blanking interval" in depth, allows both electrical and acoustic transients from the high-power ping to die down. The depth range instantaneously ensonified by a given pulse is called the "pulse length". For a 150-kHz instrument, typical pulse and bin lengths are 8 m, and a minimal blanking interval is 4 m. These parameters can be set in the data acquisition software. The center of the first depth bin--where the first velocity estimate can be made--is the transducer depth plus the blanking interval plus half the sum of the pulse and bin lengths (which do not have to be equal).



The first generation of commercial ADCPs, most commonly from RDI, use the simplest possible pulse, or ping: a pure tone that is switched on for a few milliseconds and then switched off. These instruments are described by RDI as "narrowband" because the bandwidth of the pulse is as narrow as it can be given the length of the pulse. Bandwidth is inversely proportional to pulse length. Estimation of the Doppler shift of the backscattered sound is straightforward and unambiguous; it is based on the covariance of the signal with itself, with a very short lag.


The second generation of RDI sonars uses "coded pulses", and is described as "broadband". (See Pinkel and Smith, 1992, for a thorough description of coded pulse sonar.) Each pulse consists of several repetitions of a binary pseudorandom code sequence; each code sequence consists of several bits (e.g., 13); and each bit has the same duration, with ones are distinguished from zeros by a 180-degree reversal in the phase of the tone. Because the phase of the tone is reversed many times during the pulse, at intervals as short as the bit length, the bandwidth is inversely proportional to the bit length, not to the much longer pulse length. Hence, the bandwidth is much broader than for an uncoded pulse. With coded pulses, estimation of the Doppler shift involves the complex covariance of the signal with itself, with a lag equal to the code length--much longer than the lag used to estimate the Doppler shift with an uncoded pulse. What is actually measured is the phase change caused associated with the Doppler shift--caused by the relative motion of the scatterer--during this lag. Because the measurement is one of phase, it is inherently ambiguous. Positive 270 degrees, for example, is indistinguishable from -90 degrees. Only if the velocity is small enough to guarantee phase shifts in the range +-180 degrees will the estimate of velocity be unambiguous. The velocity corresponding to a 180 degree phase shift is therefore called the "ambiguity velocity". The longer the code (and hence the lag used in the covariance calculation), the smaller is the ambiguity velocity.


Coding can be thought of as increasing the amount of information in the pulse--hence the requirement for broader bandwidth--which in turn increases the potential accuracy of the velocity determination for a given bin length, or allows finer vertical resolution for a given accuracy. The penalty is reduced range; with increased signal bandwidth comes increased noise bandwidth, and therefore a higher noise level overall. Signal plus noise within the receiver bandwidth is the "received signal strength", sometimes called simply "amplitude" or "echo intensity"; it is recorded by the instrument on an arbitrary integer scale from 0 to 255, with each count corresponding to about 0.45 db. An estimate of signal as opposed to noise is the magnitude of the correlation--the phase of the correlation gives the velocity estimate. Correlation magnitude is also measured on a scale of 0 to 255, with 255 presumably being 100% correlation. Typical near-surface values obtained from instruments in the field may be half this amount.


Coded pulse sonars have more free parameters to be set than do uncoded pulse sonars. Given the same pulse and bin lengths, the bit length (hence bandwidth) and code length (hence lag and ambiguity interval) can be varied to optimize the tradeoff between range and accuracy. RDI provides three bandwidth settings: the default and broadest is WB0 (25%), with a maximum effective ambiguity interval of about 7.0 m/s in radial velocity component; WB1 (12%), with maximum ambiguity velocity about 3.3 m/s; and WB2 (3%), with maximum ambiguity velocity about 1.65 m/s. (These numbers are from RDI FSB-119, and I suspect they may include errors. Parts of FSB-119 are confusing and misleading at best. Errors and lack of clarity have been consistent features of RDI's documentation of their key broadband parameters and performance.) For a shipboard system, WB2 is useless; the ambiguity velocity is simply too small. RDI sometimes recommends WB1, arguing that 3.3 m/s is adequate for typical research ship speeds, given a 45-degree transducer orientation. This argument is misleading because it neglects the contribution of pitch and heave (short-term vertical velocities of the transducer) to the axial velocity component. In practice, axial velocities often will exceed 3.3 m/s, so WB0 is the only reasonable choice for routine use underway; and there is nothing to be gained by using less than the maximum ambiguity velocity. (Note that although one can request a specific ambiguity velocity with the WV command, the lag, and hence the actual ambiguity velocity, changes in fairly large discrete steps. The only way to find out what the ambiguity velocity actually is with a given WV setting is to calculate it from the lag recorded in the raw data.)


For many years, RDI attempted to get the best of both worlds--high accuracy and/or greater range together with a large dynamic range of velocity--by using schemes to resolve the velocity ambiguity, thereby allowing longer lags than would otherwise be possible. The schemes of primary interest are what they call modes 1, 4, and 7 (WM1, WM4, and WM7). Mode 1 uses no ambiguity resolution from ping to ping, but does cross the ambiguity boundary within a single ping. It is now the standard mode. Although it was originally suggested by RDI that mode 4 could be used for shipboard systems, it was hopeless from the start--a recipe for utterly useless numbers. Mode 7 was specifically designed for shipboard use, and was much more promising. It combines relatively narrow bandwidth, for range, with a short-lag "pre-ping" before each main ping to resolve the ambiguity. Under ideal conditions, it works, and it does greatly improve the range; but unfortunately it is unreliable, and no longer recommended.


Acknowledgments


Darryl Symonds of RDI was very helpful throughout the testing project; answering questions, reviewing data (particularly for the BB300), and suggesting troubleshooting strategies. Bill Hamberg extensively reconfigured the PCs and wiring, operations critical to the success of the trials. The USCG marine science technicians, and Eldridge McFadden in particular, have been consistently helpful and professional in the face of overwhelming responsibility. Chief Scientists Lisa Clough and Kelly Falkner ably juggled the needs of all participants, ensuring successful and pleasant cruises. Dale Chayes provided valuable information and advice concerning communications and the Science Data network; John Freitag was instrumental in getting us involved in this project, as well as assisting with technical work on the ship, particularly with the gyro interface.


Figures


Fig. 1: Approximately two hours of BB150 ADCP data collected and averaged by Transect on the cold water trials (leg1) is shown here in plan view (top panel) and as a color contour for eastward and northward velocities. Several features are typical of data collected during the Healy trials: (1) In the top depth bin, the anomalous velocity in the direction of ship's motion in the top bin (in this case, northward and westward) are caused by ringing. (2) Depth penetration over the two-hour segment varies from 130m to 180m: factors such as local ocean acoustic scattering levels, sea state, and heading contribute to the variability in effective depth penetration.


Fig. 2: Minimum correlation among all four beams is colored as a function of depth during a test wherein speed was incremented from 1 to 16 kts over the course of nearly 5 hours. The ship was held at each speed for about 20 minutes; data collection parameters were sometimes changed during a constant-speed interval. Five to twenty minutes of continuous data collection were available at each speed. These chunks of data are shown here separated by light blue lines that represent intervals of 5-20 minutes. The white line and numbers show the ship's speed in knots. On a ping-by-ping basis, the effective range (black circles) can be estimated as the shallowest depth where the minimum correlation falls below 64. Processed data are usually averaged over 5-minute ensembles, during which some minimum percentage of pings must be good in each depth bin for which a valid average is obtained. The effective range (black line) for processed data is therefore the point where the percent good falls below a threshold, chosen here as 80%.


Fig. 3: The three highest ship speeds from Figure 2 are shown here in order to better see the effects of bubbles at higher speeds. The heavy black line is a scaled heading of the ship

3*(H - HMEAN)). At 13 kts, five bubble events are seen, each sampled by a single ping. At 14 kts, the time interval between events is similar, but the events are occurring with a longer duration (3-8 pings). The low correlation bubble events are obviously associated with times when the black line slopes down to the right (the ship is moving most rapidly to port). At 16 kts, individual events are difficult to distinguish, and very little data is valid below 5 bins.


Fig. 4: Signal strength from beam one, for the 16-kt segment from Figures 2 and 3. The layer of bubbles under the transducer causes a decrease in amplitude at a given depth (in this test at ship speeds under 8kts, this effect was overshadowed by ocean scattering variability, (not shown)), a very slight increase in the general noise floor (not perceptible in this plot), and single pings or groups of pings with a slightly lower noise floor and extremely weak signal strength even in the top bin (for example, pings 1, 9, and a group of 3 near the center of the time range).


Fig. 5: Beam 1 correlation (BB-150) during tests of interference with other sonars. The Knudsen 3.5kHz sub-bottom profiling chirp was the most obvious in the BB150 signal (Panels 1 and 3). It should not affect the velocities since the chirps have a low correlation. The 350 kHz speedlog and the SeaBeam showed no interference in the BB150 signal (Panel 2). The BB300 pings were obvious below the depth of valid returns (Panel 4). It should be noted that although these acoustic interference signals did not affect the data during this test, we were nearly stationary and in relatively calm seas. In another instance, when steaming over water 300m deep, velocities in occasional pings and bins were anomalous, possibly due to a combination of acoustic interference and reverberation between the bottom and the ocean's surface. A median filter of the velocities prior to averaging would eliminate this problem.


Fig. 6: Single-ping velocities measured by the BB150 in the direction of the ship's heading, and referenced to the average of bins 3-8. Ensembles for each speed are offset horizontally in 1-m/s increments, and vertically by the ship's speed in knots. Ringing is obvious (a bias towards zero (left) in the top bin) for ship speeds above 11 kts, and more subtle at lower speeds.


Fig. 7: Beam 1 correlation, showing the unidentified interference we call "hash", which was present intermittently during both warm and cold water trials. Characteristics include: (1) high correlation and signal strength in all beams at an isolated bin; (2) the signal appears at intervals of 3 bins, or roughly once every 0.037 sec. In the example shown here, the hash stopped when the speed dropped, shortly after 9315. The few subsequent correlation glitches are from acoustic interference by other sonars.


Fig. 8: Minimum correlation among all four beams, showing the effect of hash and bubbles. This segment of data lasted 20 hours and included (A) a slow survey at 2kts, (B) a Seabeam canyon survey at 12kts, and (C) steaming at 16 kts. Samples at each speed are shown. Hash can be seen in the first panel (A) as regular high correlations below 200m. Hash is less clear in panels B and C, but is present at lower correlations. Bubbles are evident in panel B as the three vertical stripes of low correlation between 50 and 150m. Bubbles obliterate many pings in panel C, reducing the correlation to very low values.


Fig. 9: Signal strength from a segment of the data sampled by Panel (C) of figure 8, showing the effect of bubbles and hash together. Compare to Figure 4, which also shows the effect of bubbles at 16 kts, but contains no hash. The hash in Figure 9 is seen as anomalously high signal strengths below bin 10, often occurring every third bin in the vertical in a given ping. The bubbles cause the generally low signal strength, with virtually no signal at all in some pings and short groups of pings. Note that in the pings when signal strength is low throughout the water column, the hash is still visible close to the surface; it seems unaffected by the bubbles. This indicates that the hash is not acoustic noise coming from outside the ship.


Fig. 10: Histograms of single-beam velocity from the hour-long samples (A,B, C) in Figure 8. These velocities were screened only to reject minimum correlations under 64. The number of points available in (C) is far fewer than in (A) or (B), because of the effect of the bubbles. Note that the beam velocities in panel (A) show a bimodal distribution. The correct velocity peak is the large red hump. The second hump (to the left, or negative side of the main hump) consists entirely of data in the hash region: these data were included because their anomalously high correlations did not get screened out. Opposite pairs of beam velocities from the negative hump will yield high vertical velocities; but the vertical velocity estimates from orthogonal beam pairs are often similar, so the error velocity can be low. The bridge connecting the two humps shows the bias in the transition region between shallower bins where there is a valid signal from the water, and the deeper bins where the only signal is from the hash.


Fig. 11: Histograms of Panel A beam velocities from Figure 10 are shown with increasingly stringent editing criteria applied. The red lines in this figure are the same as the red lines in Figure 10: histograms of beam velocities, one per beam, with correlations over 64. If velocities with an absolute value of error velocity greater than 0.1 m/s (a very stringent criterion) are rejected, many points from the second velocity hump are removed and the number of points in the main hump also decreases. If data are further screened to reject absolute values of vertical velocity greater than 1.0m/s, the result is the small circles. The data from the second hump is now gone, but about half of the points in the main hump are also lost.


Fig. 12: The top panel shows the speed of the ship over ground computed from navigation (red) and the measured flow past the ship (black circles). Each one of the circles near zero corresponds to time when one or more of the beam velocities exceeded the ambiguity interval, 3.3 m/s. From timestamp 7690-7900 the ship's speed was below 14 kts, but there are a significant number of ambiguity wraps. From timestamp 8000-8150 the ship's speed ranged from 11-12kts and in this region there was still on average one wrap every 7 minutes over 130 minutes. These wraps show up in both w and the error velocity (second panel). The bottom right four panels show the individual beam velocities for bin 5. Beam 1 shows the highest level of wrapping in part because of its orientation relative to the keel, and in part because of its vertical alignment. The bottom left panel shows the vertical velocity plotted as a function of error velocity. Note that an error velocity editing criterion would eliminate most of the wrapped points, but there would still be a little cluster at +3m/s requiring editing with a vertical velocity threshold.